Keywords: Adaptive and Learning Systems
Abstract: Autonomous mobile robotic systems have been widely used in real-world applications, from environmental monitoring tasks to tasks where robots assist humans in social settings. Motion control is fundamental to autonomous mobile robotic systems, and the methods have evolved from model-based methods (that rely on physical models of the robot and its operating environment) to machine learning based methods (that use data generated by the robot while interacting with the environment).

In this talk, I will discuss both model-based control and learning-based control, and present their applications in controlling autonomous surface vehicles for dynamic plume tracking and controlling autonomous ground vehicles for pedestrian flow optimization. Challenges in controlling mobile manipulators for fast-speed sports will also be presented. Advantage and limitation of different methods will be discussed, and insights toward integrating physics-based methods with data-based methods will be provided.

Keywords: Adaptive and Learning Systems, Optimal Control, Control Applications
Abstract: Buildings continue to represent one of the largest sectors for electricity consumption globally. A significant portion of this demand is driven by heating, ventilation, and air conditioning (HVAC) systems within these structures. Due to the complexity associated with modeling large-scale HVAC systems, traditional model-based optimal control strategies become increasingly impractical. In this work, we present a game-theoretic approach to optimal control for building HVAC systems, framing the problem as a two-player non-zero-sum cooperative game. We propose a data-driven, model-free state feedback Q-learning value iteration method that addresses the quadratic game optimization problem without requiring any prior knowledge of the zone's dynamics. Mass flow rate and supply air temperature are treated as the two primary decision-making players. The building’s HVAC zone is considered as an environment in which these players interact, with its underlying dynamics remaining entirely unknown to them. The Q-learning value iteration algorithm is demonstrated to effectively learn optimal game policies for both players using input-state data under external disturbances, notably without the requirement of an initially admissible policy—a key advantage in scenarios with limited prior information on dynamics. The convergence of the proposed value iteration algorithm to the Nash equilibrium is formally established. Numerical results validate the effectiveness of the proposed approach in maintaining temperature regulation, even in the presence of unknown zone behavior and external disturbances.

Keywords: Assistive and Rehabilitation Robotics, Biomechanical Systems, Control Applications
Abstract: Stroke, spinal cord injuries, and other neurological conditions may cause neuromotor impairment from lasting symptoms that require occupational and physical therapies. Allowing for high repetitions and innovative assistive methods, robotic exoskeletons are pushing the boundaries of rehabilitation to assist in rehabilitative therapies. Deep neural networks (DNNs) have been shown to out-perform neural networks for real-time control applications, especially with complex model dynamics and allow for accurate model approximation. Used with highly transparent exoskeletons, DNNs have been shown to increase trajectory tracking performance towards an increased repetition count, which has been proven to increase movement functionality through rehabilitation. Alongside robotic therapies, functional electrical stimulation (FES) is known to recruit muscle fibers and assist individual with the addition of human movement into the task. However, FES contributes non-linear model dynamics due to aspects such as fatigue or neural spasticity from neurological conditions. Utilizing DNN based control algorithms, we explore the simulated effects of exoskeleton assistance with FES for controlled reference trajectory tracking task for a single degree of freedom. This poster showcases simulated results, accuracy, and variance of an upper-extremity trajectory tracking wearing an exoskeleton fused with FES assistance via DNN function estimation used to decrease model non-linearities.

Keywords: Biomechanical Systems, Sensors and Actuators, Mechatronic Systems
Abstract: Anterior cruciate ligament (ACL) injuries are prevalent in athletes during high-acceleration movement. This particularly affects female athletes due to a combination of biomechanical, anatomical, and external factors including hormone fluctuations, quadricep dominance, and differences in neuromuscular activation patterns. This study aims to develop a computational model to better understand the biomechanics of knee loading.

The focus is on analyzing biometric data and neuromuscular activation patterns during dynamic movements using a sensory system comprised of electromyography (EMG) sensors and pressure insoles. The EMG sensors capture muscle activation signals, providing insight into the timing and intensity of contractions involved in knee loading. Seven sensors were adhered to each leg along relevant muscles. Muscle action potential data is transmitted wirelessly via Bluetooth to centralized receivers. The data is streamed in real time and collected using a MATLAB script. Simultaneously, pressure-sensing insoles record the location and magnitude of ground contact forces. Together, these systems enable synchronized analysis of muscular and mechanical loading during dynamic activity. In this poster, details of the experimental setup will be presented along with example experimental data and preliminary modeling results.

Keywords: Control Applications
Abstract: The COVID-19 pandemic revealed major gaps in forecasting epidemic spread, particularly in estimating basic and effective reproduction numbers (R0, Re). We propose a novel computational framework to estimate these metrics from real-time data using a mechanistic, spatiotemporal PDE-based epidemic model. Applied to COVID-19 data from Hamilton County, Ohio, the predicted Re values closely align with actual spread trends across three distinct periods. They also match independent estimates from the Wallinga-Teunis and Cori methods. The results validate the framework’s accuracy and highlight its potential for future epidemic monitoring, even under sparse and evolving data conditions.

Keywords: Cyber physical systems, Discrete Event Dynamic Systems, Uncertain Systems and Robust Control
Abstract: In this paper, we present a robust framework for learning Linear Temporal Logic (LTL) formulas from positive and negative system demonstrations with uncertain measurements.

State-of-the-art inference methods for LTL formulas assume that these demonstrations are not corrupted, e.g., by noise. Our uncertainty-aware framework to learn LTL formulas includes the uncertainty information to infer an LTL formula. We model uncertainty via groups of demonstration estimates and enforce group-level correctness by requiring at least one trace per group to satisfy the learned formula. Incorporating prior knowledge further guides learning, enabling accurate and interpretable extraction of LTL for safety-critical systems.

Keywords: Estimation, Healthcare systems, Modelling and Control of Biomedical Systems
Abstract: We developed a blood pressure prediction algorithm for critically ill subjects receiving blood transfusion using a population-informed sequential inference approach. A key challenge is the conditional observability of the system dynamics, i.e., it is fully or partially observable depending on both input (hemorrhage and transfusion rates) and state variables. To predict blood pressure irrespective of the challenge, we developed an algorithm consisting of 3 recursive estimators: an extended Kalman filter (EKF) capable of inferring 4 states and two recursive parameter estimators capable of inferring 2 states and 1 state, respectively. At each measurement instant, the algorithm chooses a recursive estimator compatible to the degree of observability at that instant and infers state variable(s), and predicts blood pressure by simulating the system dynamics into the future. Initial development of the algorithm based on an in vivo large animal dataset demonstrated the proof-of-principle of the blood pressure prediction algorithm.

Keywords: Estimation, Machine Learning in modeling, estimation, and control, Agricultural Systems
Abstract: This work focuses on an approach for state and parameter estimation of a soil-moisture monitoring system consisting of an above-ground transmit coil inductively coupled to a buried passive sensor. The system is modeled as a pair of coupled resistor-inductor-capacitor (RLC) circuits and a dual extended Kalman filter (DEKF) approach is developed to simultaneously estimate the states and unknown parameters of the system. A Bayesian estimator in the form of a particle filter is designed to estimate the location of the buried sensor relative to the transmit coil based on the mutual inductance inferred by the DEKF. The localization process improves the spatial alignment between the transmit coil and the buried sensor, thus strengthening the inductive coupling effects to enhance parameter estimation. By estimating the self-capacitance parameter of the buried sensor, the moisture levels of the surrounding soil can be determined. The proposed approach is validated in simulation and physical experiments, where successful buried-sensor localization and estimation of soil moisture level are shown. These initial results can be used to design an autonomous irrigation system for precision agriculture to regulate soil-moisture level.

Keywords: Human-Machine and Human-Robot Systems, Adaptive and Learning Systems, Robotics
Abstract: Human-robot collaboration has the potential to combine the strengths of humans and robots to enhance productivity, flexibility, and safety. Achieving this potential requires a system where knowledge of relevant human and robot states can be shared with a task planner to optimize performance. A promising approach to obtaining this knowledge is through digital twins, which use prior knowledge of an agent through models while adapting to changes using real-time data. When embedded in an appropriate framework, digital twins can estimate and predict states, enabling more informed and adaptive task planning. In this work, we implement a system of digital twins in which state information from a human-robot team is used to inform task reallocation in response to anomalies, with the goal of maximizing overall team throughput. The case study for this work involves an assembly task in which three humans (one real, two virtual) work with a dual-arm robot. Humans are primarily responsible for performing assembly operations, while the robot delivers parts and can assist with select assembly steps. In aggregate, the digital twins predict the throughput of the human-robot team and detect anomalies such as when predicted throughput drops below a threshold. When an anomaly is detected, the task planner reallocates tasks among the agents based on their estimated capacity to maintain performance. The study examines how task reallocation decisions and overall throughput are affected by varying a threshold on digital twin uncertainty, such that anomalies are ignored if the associated uncertainty is too high. Additionally, we examine how incorporating digital twins that estimate human physiological states, specifically stress, affects task planning outcomes and throughput. Findings from this study aim to inform the design of future digital twin systems that support adaptive and productive human-robot collaboration.

Keywords: Human-Machine and Human-Robot Systems, Robotics, Sensors and Actuators
Abstract: Involuntary head motions elicited when a person experiences a mobile robot's point-of-view in virtual reality (VR) can serve as an additional "human-as-sensor" cue for state estimation. In a pilot study, a participant watched stereo video captured on a 4-wheel, differential ground robot using a VR headset instrumented with an inertial-measurement unit. Synchronized analysis of the baseline (unaltered-video) condition revealed a moderate negative correlation between the participant's head and the robot's pitch angles, with weaker negative correlations in roll and yaw - confirming earlier assumption that visual stimuli alone triggers vestibular-like responses.

Building on these results, we propose a second guided-viewing condition in which an on-screen horizon/heading guide encourages the viewer to align their head with the robot's body-fixed frame . We hypothesize that this lightweight visual prompt will strengthen human-head-robot coupling across all three axes by providing an explicit orientation reference, producing a higher-fidelity, low-latency orientation estimate derived from the human vestibulo-ocular response and potentially back-up on-board inertial odometry.

The poster will outline the experimental design, anticipated analysis pipeline, and avenues for fusing human-derived orientation cues with conventional sensors. By leveraging natural, involuntary head movements - and gently nudging them with minimal visual overlays - we aim to demonstrate scalable, human-sensing-in-the-loop perception that improves robot attitude estimation without adding hardware to the platform and backing up inertial odometry.

Keywords: Intelligent Autonomous Vehicles, Sensors and Actuators, Security and Privacy
Abstract: Sensor fusion plays a crucial role in ensuring robust perception and decision-making in autonomous vehicles (AVs). This work presents a lightweight sensor fusion framework that integrates 2D LiDAR with YOLOv5-based object detection to enable real-time, way-point-independent navigation. The system is experimentally validated on a scaled autonomous vehicle platform, demonstrating consistent stop sign detection and reliable stopping behavior across different experimental runs, even under slight variations in detection distance and environmental conditions. Furthermore, the framework is evaluated under adversarial image attacks using perturbed stop signs, where it maintains functionality with minor reductions in detection range and stopping precision. The presented results are validated on the Quanser QCar platform, which confirm the effectiveness of sensor fusion for real-time AV navigation while highlighting key vulnerabilities to adversarial perturbations.

Keywords: Machine Learning in modeling, estimation, and control, Power and Energy Systems, Cyber physical systems
Abstract: Accurate and cost-efficient monitoring of water levels and energy usage in water treatment systems is essential for the reliable and energy efficient operation of water supply infrastructure. Conventional estimation approaches for water level measurements either rely on sensor redundancy to compensate for uncertainty, or on physics-based models that depend heavily on the resolution and reliability of sensor inputs. Both approaches increase system cost and are sensitive to noise and unknown dynamics. This work develops an economic water level monitoring system for interconnected tanks, integrating direct measurements from ultrasonic sensors and flow meters. A supervised machine learning framework is introduced to enhance volume estimation accuracy, trained on a custom-labeled dataset built under different operating conditions. Experimental validation shows that learning-based estimators, particularly Support Vector Machines, can be comparable to traditional approaches in terms of robustness and accuracy Predicted water levels can then be used to forecast the system energy usage. The resulting system offers a scalable, affordable, and resilient solution for smart water monitoring, reducing the reliance on high-specification sensors or complex sensor architectures. The proposed energy prediction approach also enables energy-aware decision making of the water treatment system, particularly in environments where cost, simplicity, and adaptability are operational priorities.

Keywords: Machine Learning in modeling, estimation, and control, Power and Energy Systems, Estimation
Abstract: Accurate estimation of internal resistance (IR) is crucial for effective health monitoring of lithium-ion batteries (LiBs). This work presents a comparative analysis of deep learning models, Long Short-Term Memory (LSTM) networks and two variants of Physics-Informed Neural Networks(PINNs)-for IR estimation using real-world cycling data. The LSTM model captures temporal dependencies in battery behavior but lacks physical interpretability. In contrast, PINNs embed electrochemical principles into the learning process, improving generalization and physical consistency. We evaluate two PINN formulations: a physics-based PINN-Verhulst model using a logistic degradation function, and a data-driven PINN-DeepHPM model that learns latent physical dynamics. Our results using real-world battery data show that PINN-DeepHPM achieves the best performance, with a Root Mean Squared Error (RMSE) of 0.0233, Root Mean Squared Percentage Error (RMSPE) of 0.0813, and Mean Absolute Percentage Error (MAPE) of 4.462. This is followed by the PINN-Verhulst and LSTM models with comparatively higher error metrics. These findings demonstrate that integrating learned physics significantly enhances prediction accuracy and robustness, making PINN-DeepHPM a strong candidate for practical battery diagnostics.

Keywords: Modeling and Validation, Power and Energy Systems
Abstract: Underground mining operations use electric equipment attached to high-voltage cables, which could be dangerous to the personnel who work around them. Several mines use diesel equipment that significantly increases flexibility of operations. However, they emit diesel particulate matter (DPM) that is known to be carcinogenic. Therefore, there is an increased interest in using large-format lithium-ion battery (LIB) powered equipment underground for their high energy density, negligible emissions, and low noise and heat footprint.

Although safe under nominal operation conditions, LIBs have been reported to undergo rapid failure through a process called thermal runaway (TR). This is a major challenge for their large-scale implementation, especially in large format configuration. TR can be initiated due to mechanical damage to the LIB, overcharging, overheating, or aging-related Columbic inefficiency. LIB TR presents a hazardous condition due to the rapid release of combustion products, including toxic and inflammable gases that can rapidly spread throughout the mine as a result of ventilation airflows.

In this context, we propose using a deep learning approach to rapidly predict the thermal impacts of a TR event in an underground mine tunnel. Specifically, we propose to use the `Next Generation Reservoir Computing (NGRC)' framework to reliably capture the transient-state thermofluid parameters.

Keywords: Modelling and Control of Biomedical Systems, Healthcare systems
Abstract: This work examines the problem of safety control of automated propofol infusion, a relevant problem given its safety-critical nature. The literature presents many closed-loop controller designs for this problem, but there is little work on the application of control barrier functions (CBFs). Moreover, this problem has relative degree 2, which requires additional mathematical conditions for safety. To explore this we (1) demonstrate the inefficacy of a typical second-order exponential CBF and (2) implement a nonlinear “graceful” CBF that respects safety constraints.

Keywords: Path Planning and Motion Control, Machine Learning in modeling, estimation, and control, Robotics
Abstract: When a snow avalanche involves buried human victims, a victim's survival rate is critically dependent on search and rescue time. In fact, if a victim is found and uncovered within 15 minutes of being buried, they have over a 90% chance of survival. This work focuses on developing a search algorithm that leverages ergodic exploration and sequential Monte Carlo estimation to improve the performance relative to traditional methods that follow the magnetic field gradient measured by an avalanche transceiver. First, a sequential Monte Carlo estimator is utilized to estimate the victim's pose (location and orientation). The machine learning approach estimates the parameters of a magnetic dipole model of the victim's avalanche transceiver in transmit mode. Next, the concept of ergodicity is leveraged to take the posterior of the estimator to create search trajectories that maximize information gained to improve both the parameter estimation process and the localization of the buried victim. Physical experiments are performed with off-the-shelf avalanche transceivers and an autonomous quadruped robot to validate the approach. It is demonstrated that ergodic exploration significantly increases the convergence rate and accuracy of the victim search process compared to traditional search.

Keywords: Power and Energy Systems, Transportation Systems, Underwater Vehicles
Abstract: This work examines the problem of maximizing the surge performance of a wave glider over a broad range of sea states. A wave glider is a passive energy harvesting device consisting of a floating body attached to a winged underwater glider through an elastic tether. As the float heaves (i.e., rises and falls) with incoming waves, it induces glider heave motion through the tether. This, in turn, causes the glider’s hydrofoils (or “wings”) to experience nonzero velocities relative to the surrounding fluid. The hydrodynamic wing forces produced by these relative velocities induce surge (or forward) motion of the entire system. This allows wave gliders to exploit incoming wave energy, purely passively, to induce forward surge motion. There is a rich existing literature on wave glider modeling. However, the problem of utilizing the resulting models for design optimization and sensitivity analyses remains relatively unexplored. We address this gap by building a multibody physics-based model of the glider system’s dynamics, and utilizing it to explore the impact of one particularly critical design parameter (namely, tether stiffness) on overall surge performance. The model captures the dynamics of: float heave and surge; glider heave, surge, and pitch; hydrofoil pitch; tether deflection; and system hydrodynamics. As shown in the work, the choice of tether stiffness has a very significant impact on overall system surge performance. In particular, stiffer tethers tend to translate into forward surge velocities that are both higher and more robust to the choice of hydrofoil return spring stiffness for any given sea state. These results provide valuable initial insights for potential future work on wave glider system design optimization for surge performance.

Keywords: Power and Energy Systems, Underwater Vehicles
Abstract: This work presents a dynamic cost analysis for assessing the sustainability of mobile, wave-powered desalination systems using a Dual Markov Chain approach. Specifically, it models a wave glider-based platform using a stochastic aging framework to evaluate long-term economic viability.

Keywords: Power and Energy Systems, Unmanned Ground and Aerial Vehicles, Optimal Control
Abstract: This work simultaneously optimizes the design and control of an unmanned aerial vehicle (UAV) for endurance (i.e., the ability to execute missions reliably over large distances and time durations). The work focuses on a tiltrotor UAV that combines the flexibility of vertical takeoff and landing with the efficiency of horizontal cruise. A parallel hybrid powertrain enables this UAV to utilize an energy-dense combustion subsystem during cruise, while meeting the significant additional power demands during takeoff and landing using a power- dense electric subsystem. There literature already provides extensive models of this powetrain’s components (i.e., engine, motor/generator, lithium-ion battery, etc.). We use these models for endurance maximization. This is a challenging stochastic optimization problem, because it is impossible to predict exactly how much power is needed versus time for a given mission due to uncertainties such as wind, changing mission tasks, and other random events. To address this challenge, we build a detailed UAV mission simulator, and use it to perform a Monte Carlo simulation of a stochastic family of missions. We then pose a control co-design problem where the objective is stochastic endurance maximization, and the optimization variables include the sizing of the various powertrain components as well as the overall power management policy. This policy dictates the ratio of electric power to mechanical power in terms of a weighted summation of Gaussian kernel functions of fuel availability, battery state of charge (SOC), and total power demand. The resulting co-design framework is valuable not just for stochastic endurance maximization, but also for gaining insights into the impact of component (e.g., battery) performance limitations on achievable UAV endurance.

Keywords: Modelling, Identification and Signal Processing, Estimation, Automotive Systems

Keywords: Assistive and Rehabilitation Robotics, Biomechanical Systems, Human-Machine and Human-Robot Systems

Keywords: Assistive and Rehabilitation Robotics, Control Applications
Abstract: Robotic exoskeletons provide a promising approach into improving traditional stroke rehabilitation with unique interactions and sensing modalities. In this manuscript, we explore the use of Deep Neural Networks (DNNs) as function estimators for any un-modeled dynamics especially in highly non-linear system. Using Lyapunov stability theory, the Lyapunov-based Gradient Descent (L-GraD) controller was designed to feed a desired reference trajectory into an impedance controlled system. Adjusting DNN weights in real-time improves tracking performance, and with the highly transparent and compliant exoskeleton, has potential for successful clinical implementation. Monte Carlo simulation results show that real-time DNNs for non-linear dynamics improve control performance and reduce mean squared error during disturbance episodes. Results indicate a DNN with three hidden layers and 15 neurons each will provide the best results while maintaining lightweight architecture. Experimental results validate this L-GraD controller with improved performance over traditional control methods.

Keywords: Biomechanical Systems, Modeling and Validation
Abstract: Marine mammals rely on their flukes for propulsion. However, the forces generated by their foil-like flukes can not be measured directly due to the complexities of the marine environment. This study presents a data-driven modeling framework to investigate propulsive hydrodynamic forces during swimming. First, synthetic data was generated using a low-order simulation based on prior research to generate training data for model identification. The simulation models the two-dimensional translational motion (longitudinal and vertical) of the animal and approximates its fluking gait as a multi-linkage system. The propulsion force acting on the fluke is simulated using the principles of unsteady hydrodynamics and hydroelasticity. Subsequently, extended dynamic mode decomposition identifies a nonlinear model by lifting the original state-space into a higher-order nonlinear representation. The results demonstrate that the proposed method accurately estimates both the motion of the animal and the hydrodynamic forces exerted on it.

Keywords: Healthcare systems, Modelling and Control of Biomedical Systems, Control Applications
Abstract: This work focuses on the efficient regulation of blood glucose (BG) in people with Type 1 Diabetes (T1D) by means of automated subcutaneous insulin delivery through a pump, which is essential to mimic the real-life functioning of a healthy pancreas and improve the quality of life for patients. In particular, this research addresses one of the bottlenecks in insulin pump therapy, namely the detection of insulin infusion occlusions and consequent interruption of the flow of insulin into the portal circulation, before the degradation of glucose control can be measured. We exploit a digital twin of a mechanical pump and incorporate it in a large scale metabolic model of T1D glucose dynamics. Our failure detection algorithms is based on the displacement of the pump’s piston after an injection of insulin. We test our approach in simulation showing promising results for the early detection of insulin flow interruptions.

Keywords: Modeling and Control of Biotechnological Systems, Distributed Parameter Systems, Chemical Process Control
Abstract: Populations that are encountered in ecology, epidemics, or biotechnology, do not only interact over time but also age over time. They are modeled as age-structured partial differential equations, where age is the space variable. Involving integrals over age, they are in fact integro-partial differential equations. Population dynamics can be observed in chemostat reactors, where harvesting is used as input to regulate population densities to track desired reference trajectories. Up to date, such models have only been considered as single input, single output systems, imposing strict constraints on reference trajectories when multiple species are involved. We present a novel model for two competing populations in a cascaded chemostat setup, allowing for an independent tracking control of each species. The control is implemented in a two degree of freedom structure based on the linearized system, with an inversion-based feedforward control and a robust feedback law. The designed control is able to track sinusoidal reference trajectories around the steady state with low error, underlining the great potential of model-based control in the chemostats.

Keywords: Modelling and Control of Biomedical Systems, Control Applications, Human-Machine and Human-Robot Systems
Abstract: This paper presents a Koopman-based data-driven modeling approach for bubble-based, acoustically-actuated microswimmers. To address challenges in capturing their nonlinear dynamics, we conducted frequency sweep experiments and identified the optimal actuation frequencies for translation and rotation. Using synchronized imaging and voltage control, we collected swimmer position and orientation data under varied input amplitudes. A lifted linear model was then constructed using the Koopman operator and extended dynamic mode decomposition (EDMD), with recursive updates for real-time adaptability. The novelty of this work lies in providing an interpretable, frequency-optimized, and experimentally grounded framework for modeling acoustically-propelled microscale robots.

Keywords: Control Design, Robotics, Underwater Vehicles
Abstract: This paper presents MPC-CDF, a new approach integrating control density functions (CDFs) within a model predictive control (MPC) framework to ensure safety-critical control in nonlinear dynamical systems. By using the dual formulation of the navigation problem, we incorporate CDFs into the MPC framework, ensuring both convergence and safety in a discrete-time setting. These density functions are endowed with a physical interpretation, where the associated measure signifies the occupancy of system trajectories. Leveraging this occupancy-based perspective, we synthesize safety-critical controllers using the proposed MPC-CDF framework. We illustrate the safety properties of this framework using a unicycle model and compare it with a control barrier function-based method. The efficacy of this approach is demonstrated in the autonomous safe navigation of an underwater vehicle, which avoids complex and arbitrary obstacles while achieving the desired level of safety.

Keywords: Adaptive and Learning Systems, Control Applications, Path Planning and Motion Control
Abstract: Many autonomous robots aimed at source-seeking are studied, and their controls designed, using unicycle modeling and formulation. This is true not only for model-based controllers, but also for model-free, real-time control methods such as extremum seeking control (ESC). In this paper, we propose a unicycle-based ESC design applicable to differential wheeled robots that: (1) is very simple design, based on one simple control-affine law, and without state integrators; (2) attenuates oscillations known to persist in ESC designs (i.e., fully stop at the source); and (3) operates in a model-free, real-time setting, tolerating environmental/sensor noise. We provide simulation and real-world robotic experimental results for fixed and moving light source seeking by a differential wheeled robot using our proposed design. Results indicate clear advantages of our proposed design when compared to the literature, including attenuation of undesired oscillations, improved convergence speed, and better handling of noise.

Keywords: Adaptive and Learning Systems, Control Design, Discrete Event Dynamic Systems
Abstract: This paper presents a data-driven framework for the polyhedral stability of discrete-time systems. To this end, data-based zonotope-based parametrized representations of closed-loop systems are leveraged, and closed-loop parameters are directly learned to enforce satisfaction of the Minkowski-Lyapunov inequality via a linear programming approach. This is in sharp contrast to standard quadratic Lyapunov approaches that boil down to semi-definite programming methods. Building on this foundation, a robust stabilizing controller is developed for systems subject to process noise bounded by the infty-norm. The presented data-driven characterization of closed-loop systems using matrix zonotopes captures the effects of process noise and facilitates the design of robust controllers that ensure stability for all closed-loop systems conformed with collected data. Compared to open-loop methods, this approach uses fewer data points while guaranteeing closed-loop stability via the Minkowski-Lyapunov inequality verification. Its effectiveness is validated by a simulation that stabilizes an unstable system.

Keywords: Adaptive and Learning Systems, Control Design, Large Scale Complex Systems
Abstract: This paper presents a novel, model-free, data-driven control synthesis technique known as dynamic mode adaptive control (DMAC) for synthesizing controllers for complex systems whose mathematical models are not suitable for classical control design. DMAC consists of a dynamics approximation module and a controller module. The dynamics approximation module is motivated by data-driven reduced-order modeling techniques and directly approximates the system’s dynamics in state-space form using a matrix version of the recursive least squares algorithm. The controller module includes an output tracking controller that utilizes sparse measurements from the system to generate the control signal. The DMAC controller design technique is demonstrated through various dynamic systems commonly found in engineering applications. A systematic sensitivity study demonstrates the robustness of DMAC with respect to its own hyperparameters and the system’s parameters.

Keywords: Adaptive and Learning Systems, Control Design, Linear Control Systems
Abstract: This paper bridges optimization and control, and presents a novel closed-loop control framework based on natural gradient descent, offering a trajectory-oriented alternative to traditional cost-function tuning. By leveraging the Fisher Information Matrix, we formulate a preconditioned gradient descent update that explicitly shapes system trajectories. We show that in sharp contrast to traditional controllers, our approach provides flexibility to shape the system’s low-level behavior. To this end, the proposed method parameterizes closed-loop dynamics in terms of stationary covariance and an unknown cost function, providing a geometric interpretation of control adjustments. We establish theoretical stability conditions. Simulation results on a rotary inverted pendulum benchmark highlight the advantages of natural gradient descent in trajectory shaping.

Keywords: Adaptive and Learning Systems, Machine Learning in modeling, estimation, and control, Cognition modeling
Abstract: Wind turbine blade inspection using drone-based imaging has emerged as a highly promising solution for scalable, low-cost monitoring of large wind farms. However, the majority of existing visual inspection methods still rely heavily on either handcrafted features or large-scale labeled datasets to train traditional deep learning models. These approaches face significant challenges in terms of adaptability, requiring time-consuming retraining or fine-tuning whenever new defect patterns, lighting conditions, or inspection angles arise. In this work, we propose a novel few-shot visual reasoning pipeline based on visual -text Retrieval-Augmented Generation (RAG) integrated with a pre-trained Vision-Language Model (VLM), designed to reduce reliance on manual labeling and enhance adaptability across blade inspection scenarios. Unlike conventional pipelines, our system does not require task-specific fine-tuning. Instead, it performs in-context few-shot reasoning by retrieving relevant visual-textual examples from a structured knowledge base and prompting the language model to reason over these alongside the current image’s description. To demonstrate its practical potential, we construct a domain-specific knowledge base including structured textual files and optional curated image-caption examples. We show that our visual-text RAG-VLM framework is able to reason about damage type and severity in a flexible, interpretable, and data-efficient manner.

Keywords: Aerospace
Abstract: A novel technique is introduced herein intended to enhance legacy interceptors by augmenting the capabilities of antiquated defense systems and thus, extending the operational lifespans of conventional missiles. In the solution proposed, an emulator connects a modern guidance law to a classical guidance system by inserting a virtual target into an engagement between an attacker and real target. By strategically computing the velocity and turn rate commands of the virtual target, a conventional missile, despite being guided solely by proportional navigation, is able to follow a trajectory generated by any guidance law and intercept the real target, as the attacker pursues the virtual target. Introducing a virtual target into an engagement then shifts the modernization of weapons systems from manufacturing hardware to redeveloping software. The feasibility of this method is demonstrated through the simulation of a modern cooperative attack strategy. This approach is in alignment with the Department of Defense's ongoing initiatives towards advancing the nation's weapons systems while managing resources.

Keywords: Aerospace, Control Design, Control Applications
Abstract: This paper presents a jerk-based control allocation framework, particularly well-suited for tiltrotor vehicles, enabling the consideration of actuator rate saturations and bandwidths. It is shown how a jerk-based allocation reformulates the otherwise nonlinear allocation problem encountered with tilt rotors into a linear one, and how the allocation task can be solved efficiently exploiting symbolic calculations and pivoted LDL-matrix decomposition.The effectiveness of the approach is demonstrated in simulations for a quad-tilt rotor configuration of an airship-drone and contrasted to a proposed simplified, purely symbolic allocation strategy.

Keywords: Aerospace, Electromechanical systems, Modeling and Validation
Abstract: The powertrain design is crucial for success in the electric multi-rotor aircraft development race. It significantly impacts the aircraft's performance, maneuverability, and safety. The sizing of all powertrain components is essential for balancing efficiency, and dynamics within its entire operational envelope. This work proposes a holistic approach to analyze the dynamics of a powertrain by considering the most relevant components, including the batteries, the electronic speed controller, the motor and the propeller. The entire powertrain is formulated as a first order system which is examined via a phase plane analysis. This facilitates the assessment of the underlying dynamics on a rotational rate of change and thrust rate of change degree for every operating condition. Moreover, the whole operational envelope of the powertrain is displayed, including the subsystems' physical constraints. This results in an accurate performance prediction and acts as an enabler for effective control strategies as well as for future design optimizations.

Keywords: Aerospace, Estimation, Control Applications
Abstract: Accurate estimation of battery state of power (SOP) is essential for the safe, efficient operation of electric vertical takeoff and landing (eVTOL) aircraft. These systems face high power demands and strict safety requirements, where inaccurate power predictions can affect critical decisions—such as whether enough power remains for a safe vertical landing. Unlike traditional electric vehicles, which emphasize energy estimation via state of charge (SOC), eVTOL applications require precise SOP estimates that reflect the complex dynamics of high C-rate discharges.

This work presents a battery estimation framework tailored for high-power, safety-critical settings. The estimator uses an extended Kalman filter (EKF) and a fourth-order equivalent circuit model (ECM) incorporating SOC, two RC pair voltages, and a dynamic resistance term (R_{rm LD}) to capture lithium depletion effects. This model has been validated under eVTOL-relevant high discharge rates (up to 8C). Empirical observability analysis confirms all states are observable. A modified EKF formulation accounts for model uncertainty via process noise terms that adapt based on operating conditions, enabling dynamic adjustment of the estimator’s reliance on the model versus measurements.

To enhance robustness, we enforce physically plausible state constraints using gain projection and calibrate the filter using the normalized innovation squared (NIS) metric to ensure consistency. Simulation studies highlight the importance of non-SOC states in SOP prediction, which must be explicitly handled in the estimator design.

We evaluate the estimator using over 1000 experimental eVTOL flight profiles covering diverse conditions and failure scenarios. These are used to assess SOP prediction performance, especially in estimating hover endurance near the landing decision point—when a pilot commits to a vertical landing. Results show the EKF-based estimator improves SOP accuracy by over 60% compared to open-loop predictions. Crucially, we find that tuning solely for SOC accuracy can be misleading; optimal performance depends on downstream SOP tasks, which are sensitive to overpotential dynamics and lithium depletion onset.

Keywords: Aerospace, Estimation, Stochastic Systems
Abstract: This paper presents a robust and innovative solution for ballistic missile tracking using an Interacting Multiple Model (IMM) framework that operates without prior knowledge of missile-specific parameters. The proposed unified constant axial acceleration (UCAA) model effectively handles both boost and reentry phases, while the constant velocity (CV) model captures coast phase dynamics. The directional elevation angle serves as a clever, physics-based classifier for phase transitions, and the Monte Carlo results demonstrate strong performance across all flight regimes. The method’s independence from prior missile data makes it particularly valuable for real-world threat scenarios where such information is unavailable.

Keywords: Aerospace, Nonlinear Control Systems, Control Design
Abstract: To maintain and maneuver the attitude of large satellites representative of international space stations, and for small satellites requiring agile and large-angle attitude maneuvers, actuators are required that can provide high torque output. To solve this problem, a control moment gyro (CMG) is used, which is capable of generating a higher torque than a reaction wheel (RW), which is conventionally used for attitude control. However, there are combinations of gimbal angles in CMG systems, called singularities, where torque cannot be output in a particular axis direction. The objective of this study is to construct an attitude control system for spacecraft that takes into account the singularity avoidance of CMGs. The effectiveness of proposed method is verified by the numerical simulation.

Keywords: Adaptive and Learning Systems, Optimal Control, Control Applications
Abstract: Buildings continue to represent one of the largest sectors for electricity consumption globally. A significant portion of this demand is driven by heating, ventilation, and air conditioning (HVAC) systems within these structures. Due to the complexity associated with modeling large-scale HVAC systems, traditional model-based optimal control strategies become increasingly impractical. In this work, we present a game-theoretic approach to optimal control for building HVAC systems, framing the problem as a two-player non-zero-sum cooperative game. We propose a data-driven, model-free state feedback Q-learning value iteration method that addresses the quadratic game optimization problem without requiring any prior knowledge of the zone's dynamics. Mass flow rate and supply air temperature are treated as the two primary decision-making players. The building’s HVAC zone is considered as an environment in which these players interact, with its underlying dynamics remaining entirely unknown to them. The Q-learning value iteration algorithm is demonstrated to effectively learn optimal game policies for both players using input-state data under external disturbances, notably without the requirement of an initially admissible policy—a key advantage in scenarios with limited prior information on dynamics. The convergence of the proposed value iteration algorithm to the Nash equilibrium is formally established. Numerical results validate the effectiveness of the proposed approach in maintaining temperature regulation, even in the presence of unknown zone behavior and external disturbances.

Keywords: Assistive and Rehabilitation Robotics, Biomechanical Systems, Control Applications
Abstract: Stroke, spinal cord injuries, and other neurological conditions may cause neuromotor impairment from lasting symptoms that require occupational and physical therapies. Allowing for high repetitions and innovative assistive methods, robotic exoskeletons are pushing the boundaries of rehabilitation to assist in rehabilitative therapies. Deep neural networks (DNNs) have been shown to out-perform neural networks for real-time control applications, especially with complex model dynamics and allow for accurate model approximation. Used with highly transparent exoskeletons, DNNs have been shown to increase trajectory tracking performance towards an increased repetition count, which has been proven to increase movement functionality through rehabilitation. Alongside robotic therapies, functional electrical stimulation (FES) is known to recruit muscle fibers and assist individual with the addition of human movement into the task. However, FES contributes non-linear model dynamics due to aspects such as fatigue or neural spasticity from neurological conditions. Utilizing DNN based control algorithms, we explore the simulated effects of exoskeleton assistance with FES for controlled reference trajectory tracking task for a single degree of freedom. This poster showcases simulated results, accuracy, and variance of an upper-extremity trajectory tracking wearing an exoskeleton fused with FES assistance via DNN function estimation used to decrease model non-linearities.

Keywords: Biomechanical Systems, Sensors and Actuators, Mechatronic Systems
Abstract: Anterior cruciate ligament (ACL) injuries are prevalent in athletes during high-acceleration movement. This particularly affects female athletes due to a combination of biomechanical, anatomical, and external factors including hormone fluctuations, quadricep dominance, and differences in neuromuscular activation patterns. This study aims to develop a computational model to better understand the biomechanics of knee loading.

The focus is on analyzing biometric data and neuromuscular activation patterns during dynamic movements using a sensory system comprised of electromyography (EMG) sensors and pressure insoles. The EMG sensors capture muscle activation signals, providing insight into the timing and intensity of contractions involved in knee loading. Seven sensors were adhered to each leg along relevant muscles. Muscle action potential data is transmitted wirelessly via Bluetooth to centralized receivers. The data is streamed in real time and collected using a MATLAB script. Simultaneously, pressure-sensing insoles record the location and magnitude of ground contact forces. Together, these systems enable synchronized analysis of muscular and mechanical loading during dynamic activity. In this poster, details of the experimental setup will be presented along with example experimental data and preliminary modeling results.

Keywords: Control Applications
Abstract: The COVID-19 pandemic revealed major gaps in forecasting epidemic spread, particularly in estimating basic and effective reproduction numbers (R0, Re). We propose a novel computational framework to estimate these metrics from real-time data using a mechanistic, spatiotemporal PDE-based epidemic model. Applied to COVID-19 data from Hamilton County, Ohio, the predicted Re values closely align with actual spread trends across three distinct periods. They also match independent estimates from the Wallinga-Teunis and Cori methods. The results validate the framework’s accuracy and highlight its potential for future epidemic monitoring, even under sparse and evolving data conditions.

Keywords: Cyber physical systems, Discrete Event Dynamic Systems, Uncertain Systems and Robust Control
Abstract: In this paper, we present a robust framework for learning Linear Temporal Logic (LTL) formulas from positive and negative system demonstrations with uncertain measurements.

State-of-the-art inference methods for LTL formulas assume that these demonstrations are not corrupted, e.g., by noise. Our uncertainty-aware framework to learn LTL formulas includes the uncertainty information to infer an LTL formula. We model uncertainty via groups of demonstration estimates and enforce group-level correctness by requiring at least one trace per group to satisfy the learned formula. Incorporating prior knowledge further guides learning, enabling accurate and interpretable extraction of LTL for safety-critical systems.

Keywords: Estimation, Healthcare systems, Modelling and Control of Biomedical Systems
Abstract: We developed a blood pressure prediction algorithm for critically ill subjects receiving blood transfusion using a population-informed sequential inference approach. A key challenge is the conditional observability of the system dynamics, i.e., it is fully or partially observable depending on both input (hemorrhage and transfusion rates) and state variables. To predict blood pressure irrespective of the challenge, we developed an algorithm consisting of 3 recursive estimators: an extended Kalman filter (EKF) capable of inferring 4 states and two recursive parameter estimators capable of inferring 2 states and 1 state, respectively. At each measurement instant, the algorithm chooses a recursive estimator compatible to the degree of observability at that instant and infers state variable(s), and predicts blood pressure by simulating the system dynamics into the future. Initial development of the algorithm based on an in vivo large animal dataset demonstrated the proof-of-principle of the blood pressure prediction algorithm.

Keywords: Estimation, Machine Learning in modeling, estimation, and control, Agricultural Systems
Abstract: This work focuses on an approach for state and parameter estimation of a soil-moisture monitoring system consisting of an above-ground transmit coil inductively coupled to a buried passive sensor. The system is modeled as a pair of coupled resistor-inductor-capacitor (RLC) circuits and a dual extended Kalman filter (DEKF) approach is developed to simultaneously estimate the states and unknown parameters of the system. A Bayesian estimator in the form of a particle filter is designed to estimate the location of the buried sensor relative to the transmit coil based on the mutual inductance inferred by the DEKF. The localization process improves the spatial alignment between the transmit coil and the buried sensor, thus strengthening the inductive coupling effects to enhance parameter estimation. By estimating the self-capacitance parameter of the buried sensor, the moisture levels of the surrounding soil can be determined. The proposed approach is validated in simulation and physical experiments, where successful buried-sensor localization and estimation of soil moisture level are shown. These initial results can be used to design an autonomous irrigation system for precision agriculture to regulate soil-moisture level.

Keywords: Human-Machine and Human-Robot Systems, Adaptive and Learning Systems, Robotics
Abstract: Human-robot collaboration has the potential to combine the strengths of humans and robots to enhance productivity, flexibility, and safety. Achieving this potential requires a system where knowledge of relevant human and robot states can be shared with a task planner to optimize performance. A promising approach to obtaining this knowledge is through digital twins, which use prior knowledge of an agent through models while adapting to changes using real-time data. When embedded in an appropriate framework, digital twins can estimate and predict states, enabling more informed and adaptive task planning. In this work, we implement a system of digital twins in which state information from a human-robot team is used to inform task reallocation in response to anomalies, with the goal of maximizing overall team throughput. The case study for this work involves an assembly task in which three humans (one real, two virtual) work with a dual-arm robot. Humans are primarily responsible for performing assembly operations, while the robot delivers parts and can assist with select assembly steps. In aggregate, the digital twins predict the throughput of the human-robot team and detect anomalies such as when predicted throughput drops below a threshold. When an anomaly is detected, the task planner reallocates tasks among the agents based on their estimated capacity to maintain performance. The study examines how task reallocation decisions and overall throughput are affected by varying a threshold on digital twin uncertainty, such that anomalies are ignored if the associated uncertainty is too high. Additionally, we examine how incorporating digital twins that estimate human physiological states, specifically stress, affects task planning outcomes and throughput. Findings from this study aim to inform the design of future digital twin systems that support adaptive and productive human-robot collaboration.

Keywords: Human-Machine and Human-Robot Systems, Robotics, Sensors and Actuators
Abstract: Involuntary head motions elicited when a person experiences a mobile robot's point-of-view in virtual reality (VR) can serve as an additional "human-as-sensor" cue for state estimation. In a pilot study, a participant watched stereo video captured on a 4-wheel, differential ground robot using a VR headset instrumented with an inertial-measurement unit. Synchronized analysis of the baseline (unaltered-video) condition revealed a moderate negative correlation between the participant's head and the robot's pitch angles, with weaker negative correlations in roll and yaw - confirming earlier assumption that visual stimuli alone triggers vestibular-like responses.

Building on these results, we propose a second guided-viewing condition in which an on-screen horizon/heading guide encourages the viewer to align their head with the robot's body-fixed frame . We hypothesize that this lightweight visual prompt will strengthen human-head-robot coupling across all three axes by providing an explicit orientation reference, producing a higher-fidelity, low-latency orientation estimate derived from the human vestibulo-ocular response and potentially back-up on-board inertial odometry.

The poster will outline the experimental design, anticipated analysis pipeline, and avenues for fusing human-derived orientation cues with conventional sensors. By leveraging natural, involuntary head movements - and gently nudging them with minimal visual overlays - we aim to demonstrate scalable, human-sensing-in-the-loop perception that improves robot attitude estimation without adding hardware to the platform and backing up inertial odometry.

Keywords: Intelligent Autonomous Vehicles, Sensors and Actuators, Security and Privacy
Abstract: Sensor fusion plays a crucial role in ensuring robust perception and decision-making in autonomous vehicles (AVs). This work presents a lightweight sensor fusion framework that integrates 2D LiDAR with YOLOv5-based object detection to enable real-time, way-point-independent navigation. The system is experimentally validated on a scaled autonomous vehicle platform, demonstrating consistent stop sign detection and reliable stopping behavior across different experimental runs, even under slight variations in detection distance and environmental conditions. Furthermore, the framework is evaluated under adversarial image attacks using perturbed stop signs, where it maintains functionality with minor reductions in detection range and stopping precision. The presented results are validated on the Quanser QCar platform, which confirm the effectiveness of sensor fusion for real-time AV navigation while highlighting key vulnerabilities to adversarial perturbations.

Keywords: Machine Learning in modeling, estimation, and control, Power and Energy Systems, Cyber physical systems
Abstract: Accurate and cost-efficient monitoring of water levels and energy usage in water treatment systems is essential for the reliable and energy efficient operation of water supply infrastructure. Conventional estimation approaches for water level measurements either rely on sensor redundancy to compensate for uncertainty, or on physics-based models that depend heavily on the resolution and reliability of sensor inputs. Both approaches increase system cost and are sensitive to noise and unknown dynamics. This work develops an economic water level monitoring system for interconnected tanks, integrating direct measurements from ultrasonic sensors and flow meters. A supervised machine learning framework is introduced to enhance volume estimation accuracy, trained on a custom-labeled dataset built under different operating conditions. Experimental validation shows that learning-based estimators, particularly Support Vector Machines, can be comparable to traditional approaches in terms of robustness and accuracy Predicted water levels can then be used to forecast the system energy usage. The resulting system offers a scalable, affordable, and resilient solution for smart water monitoring, reducing the reliance on high-specification sensors or complex sensor architectures. The proposed energy prediction approach also enables energy-aware decision making of the water treatment system, particularly in environments where cost, simplicity, and adaptability are operational priorities.

Keywords: Machine Learning in modeling, estimation, and control, Power and Energy Systems, Estimation
Abstract: Accurate estimation of internal resistance (IR) is crucial for effective health monitoring of lithium-ion batteries (LiBs). This work presents a comparative analysis of deep learning models, Long Short-Term Memory (LSTM) networks and two variants of Physics-Informed Neural Networks(PINNs)-for IR estimation using real-world cycling data. The LSTM model captures temporal dependencies in battery behavior but lacks physical interpretability. In contrast, PINNs embed electrochemical principles into the learning process, improving generalization and physical consistency. We evaluate two PINN formulations: a physics-based PINN-Verhulst model using a logistic degradation function, and a data-driven PINN-DeepHPM model that learns latent physical dynamics. Our results using real-world battery data show that PINN-DeepHPM achieves the best performance, with a Root Mean Squared Error (RMSE) of 0.0233, Root Mean Squared Percentage Error (RMSPE) of 0.0813, and Mean Absolute Percentage Error (MAPE) of 4.462. This is followed by the PINN-Verhulst and LSTM models with comparatively higher error metrics. These findings demonstrate that integrating learned physics significantly enhances prediction accuracy and robustness, making PINN-DeepHPM a strong candidate for practical battery diagnostics.

Keywords: Modeling and Validation, Power and Energy Systems
Abstract: Underground mining operations use electric equipment attached to high-voltage cables, which could be dangerous to the personnel who work around them. Several mines use diesel equipment that significantly increases flexibility of operations. However, they emit diesel particulate matter (DPM) that is known to be carcinogenic. Therefore, there is an increased interest in using large-format lithium-ion battery (LIB) powered equipment underground for their high energy density, negligible emissions, and low noise and heat footprint.

Although safe under nominal operation conditions, LIBs have been reported to undergo rapid failure through a process called thermal runaway (TR). This is a major challenge for their large-scale implementation, especially in large format configuration. TR can be initiated due to mechanical damage to the LIB, overcharging, overheating, or aging-related Columbic inefficiency. LIB TR presents a hazardous condition due to the rapid release of combustion products, including toxic and inflammable gases that can rapidly spread throughout the mine as a result of ventilation airflows.

In this context, we propose using a deep learning approach to rapidly predict the thermal impacts of a TR event in an underground mine tunnel. Specifically, we propose to use the `Next Generation Reservoir Computing (NGRC)' framework to reliably capture the transient-state thermofluid parameters.

Keywords: Modelling and Control of Biomedical Systems, Healthcare systems
Abstract: This work examines the problem of safety control of automated propofol infusion, a relevant problem given its safety-critical nature. The literature presents many closed-loop controller designs for this problem, but there is little work on the application of control barrier functions (CBFs). Moreover, this problem has relative degree 2, which requires additional mathematical conditions for safety. To explore this we (1) demonstrate the inefficacy of a typical second-order exponential CBF and (2) implement a nonlinear “graceful” CBF that respects safety constraints.

Keywords: Path Planning and Motion Control, Machine Learning in modeling, estimation, and control, Robotics
Abstract: When a snow avalanche involves buried human victims, a victim's survival rate is critically dependent on search and rescue time. In fact, if a victim is found and uncovered within 15 minutes of being buried, they have over a 90% chance of survival. This work focuses on developing a search algorithm that leverages ergodic exploration and sequential Monte Carlo estimation to improve the performance relative to traditional methods that follow the magnetic field gradient measured by an avalanche transceiver. First, a sequential Monte Carlo estimator is utilized to estimate the victim's pose (location and orientation). The machine learning approach estimates the parameters of a magnetic dipole model of the victim's avalanche transceiver in transmit mode. Next, the concept of ergodicity is leveraged to take the posterior of the estimator to create search trajectories that maximize information gained to improve both the parameter estimation process and the localization of the buried victim. Physical experiments are performed with off-the-shelf avalanche transceivers and an autonomous quadruped robot to validate the approach. It is demonstrated that ergodic exploration significantly increases the convergence rate and accuracy of the victim search process compared to traditional search.

Keywords: Power and Energy Systems, Transportation Systems, Underwater Vehicles
Abstract: This work examines the problem of maximizing the surge performance of a wave glider over a broad range of sea states. A wave glider is a passive energy harvesting device consisting of a floating body attached to a winged underwater glider through an elastic tether. As the float heaves (i.e., rises and falls) with incoming waves, it induces glider heave motion through the tether. This, in turn, causes the glider’s hydrofoils (or “wings”) to experience nonzero velocities relative to the surrounding fluid. The hydrodynamic wing forces produced by these relative velocities induce surge (or forward) motion of the entire system. This allows wave gliders to exploit incoming wave energy, purely passively, to induce forward surge motion. There is a rich existing literature on wave glider modeling. However, the problem of utilizing the resulting models for design optimization and sensitivity analyses remains relatively unexplored. We address this gap by building a multibody physics-based model of the glider system’s dynamics, and utilizing it to explore the impact of one particularly critical design parameter (namely, tether stiffness) on overall surge performance. The model captures the dynamics of: float heave and surge; glider heave, surge, and pitch; hydrofoil pitch; tether deflection; and system hydrodynamics. As shown in the work, the choice of tether stiffness has a very significant impact on overall system surge performance. In particular, stiffer tethers tend to translate into forward surge velocities that are both higher and more robust to the choice of hydrofoil return spring stiffness for any given sea state. These results provide valuable initial insights for potential future work on wave glider system design optimization for surge performance.

Keywords: Power and Energy Systems, Underwater Vehicles
Abstract: This work presents a dynamic cost analysis for assessing the sustainability of mobile, wave-powered desalination systems using a Dual Markov Chain approach. Specifically, it models a wave glider-based platform using a stochastic aging framework to evaluate long-term economic viability.

Keywords: Power and Energy Systems, Unmanned Ground and Aerial Vehicles, Optimal Control
Abstract: This work simultaneously optimizes the design and control of an unmanned aerial vehicle (UAV) for endurance (i.e., the ability to execute missions reliably over large distances and time durations). The work focuses on a tiltrotor UAV that combines the flexibility of vertical takeoff and landing with the efficiency of horizontal cruise. A parallel hybrid powertrain enables this UAV to utilize an energy-dense combustion subsystem during cruise, while meeting the significant additional power demands during takeoff and landing using a power- dense electric subsystem. There literature already provides extensive models of this powetrain’s components (i.e., engine, motor/generator, lithium-ion battery, etc.). We use these models for endurance maximization. This is a challenging stochastic optimization problem, because it is impossible to predict exactly how much power is needed versus time for a given mission due to uncertainties such as wind, changing mission tasks, and other random events. To address this challenge, we build a detailed UAV mission simulator, and use it to perform a Monte Carlo simulation of a stochastic family of missions. We then pose a control co-design problem where the objective is stochastic endurance maximization, and the optimization variables include the sizing of the various powertrain components as well as the overall power management policy. This policy dictates the ratio of electric power to mechanical power in terms of a weighted summation of Gaussian kernel functions of fuel availability, battery state of charge (SOC), and total power demand. The resulting co-design framework is valuable not just for stochastic endurance maximization, but also for gaining insights into the impact of component (e.g., battery) performance limitations on achievable UAV endurance.

Keywords: Modelling, Identification and Signal Processing, Estimation, Automotive Systems

Keywords: Nonlinear Systems, Estimation, Mechatronics
Abstract: This talk presents recent results on nonlinear observers and their integrated use in modern mechatronic systems ranging from autonomous vehicles to wearable sensors. First, a new observer design technique that integrates the classical high-gain observer with a novel LPV/LMI observer to provide significant advantages compared to both methods is presented. Second, a systematic extension of the high gain observer design methodology to account for sensor noise, accommodate algebraic constraints and allow for nonlinear measurement equations is presented. Following the analytical observer results, three applications in modern mechatronic systems are discussed, including a wearable device for activity classification in Parkinson’s disease patients, autonomous cars designed for teleoperator remote intervention in the presence of large wireless communication latencies, and smart agricultural/construction vehicles that utilize inexpensive sensors for end-effector position estimation. The applications are accompanied by videos of prototype experimental demonstrations. One of these applications has been successfully commercialized through a start-up company which sells over 10,000 sensor boards each year.

Keywords: Assistive and Rehabilitation Robotics, Biomechanical Systems, Human-Machine and Human-Robot Systems
Abstract: As wearable robotic assistive devices become increasingly prevalent, it is essential to not only improve their physical performance but also deepen our understanding of how they affect the human body. This talk presents two interconnected projects that address this dual challenge. The first introduces a physics-based evaluation framework that estimates biomechanical metrics—such as joint loads, muscle forces, and fatigue—using a combination of exoskeleton sensor date and musculoskeletal modeling, eliminating the need for external motion capture or EMG systems. This integrated evaluation approach enables real-time, on-board, continuous monitoring of user-device interaction, offering a scalable solution as wearable devices grow in complexity and use.

The second project features the design and development of a novel shoulder exoskeleton that provides bilateral, decoupled assistance using a single actuator. This is achieved through a unique cam-driven, floating motor system that dynamically adjusts cable tensions to support overhead tasks at varying shoulder angles. This mechanical innovation is paired with a biomechanical study showing reduced activation in key shoulder muscles, such as the deltoids and supraspinatus.

Together, these projects demonstrate a shift in how assistive devices are designed and evaluated. Beyond providing physical assistance, future devices must incorporate intelligent sensing and modeling to understand how they are affecting the human user. By embedding evaluation tools directly into exoskeletons and validating them through novel hardware applications, this work bridges the gap between augmentation and insight—laying the foundation for next-generation wearable robotics.

Keywords: Assistive and Rehabilitation Robotics, Robotics, Control Applications
Abstract: Providing endless possibilities for control and assessment, robotic exoskeletons allow for many undiscovered human-robot interactions to further benefit movement therapies for neuromotor impairment. The nature of direct feedback and repeatability allows for intelligent control to modulate rehabilitation through improved assessment, increased repetitions, engaging training sessions, and individualized assistance. Affecting millions, neurological impairments, such as spinal cord injury, traumatic brain injuries, or cerebral vascular accidents, degrade quality of life and may lead to chronic symptoms. Impaired individuals may not have sufficient strength or motor control in regards to reaching activities and require movement rehabilitation to regain semblance of functionality. A critical role in reaching tasks and nervous system reorganization is increased repetitions with sufficient intensity of coordinated limb movements. Muscular functional electrical stimulation helps facilitate increased coordinated movements assisting during low volitional muscular control. However, these methods lead to muscular fatigue and limit the duration of rehabilitation training sessions. Combining muscular electrical stimulation with the benefits of robotic exoskeletons allow for mitigation of muscular fatigue and other unmodeled dynamics to provide increased rehabilitative exercises. Using deep neural networks to join these rehabilitative modalities allow for safe control with Lyapunov-based stability guarantees to further allow for increased coordinated movement tasks. Deep neural networks provide learned patient treatment, which targets an individualized point of care to better adapt coordinated movement therapy to each individual. Functional electrical stimulation, conjoined with an upper-extremity exoskeleton, allows for safe human robot interventions for coordinated movement therapies from tailored control methods via Lyapunov-based deep neural network control.

Keywords: Automotive Systems, Estimation, Machine Learning in modeling, estimation, and control
Abstract: Incorporating teleoperation in autonomous vehicles (AVs) will allow for effective human intervention during autonomy failures, thus ensuring safer deployment of AV technology. An important requirements for AV teleoperation is that the environment around the remote ego-vehicle needs to be recreated at the teleoperator station using camera images received over a wireless network. However, transmitting large camera and Lidar images over a wireless transmission suffers from significant problems of latency and bandwidth. In this talk a Predictive Display (PD) system which compensates for such latency and enhances AV teleoperation will be described. The PD system is based on estimating the position and orientation of the ego vehicle and of other nearby vehicles using nonlinear observers. The observer utilizes IMU, GNSS and radar sensors to perform accurate state estimation and vehicle tracking. These estimates are then used to transform delayed camera videos to create updated videos for display to the teleoperator. Image processing is done using both deep-learning methods and traditional computer vision techniques. The first part of the talk will present a human subjects study to compare teleoperation performance with and without PD. The results clearly demonstrate that even a 0.5 seconds delay in camera images can make it impossible to control the vehicle but the use of the developed PD system can enable safe remote vehicle control with almost as accurate a performance as the delay-free case. The second part of the talk elucidates the application of PD systems on real-world video data using many different algorithms. First a deep-learning based algorithm will be described which uses 3D reconstruction and image-inpainting to generate predictive video. Then a vector field based PD system will be described which uses estimated vector fields to synthesize new images from the delayed ones. The algorithms use new motion models for vehicle tracking, and novel nonlinear observers along with sensor fusion of multi-camera and Lidar data. The experimental data proves the superiority of the developed PD methods as compared to other state-of-the-art video prediction methods.

Keywords: Automotive Systems, Transportation Systems
Abstract: With the prevalence of autonomous vehicles in modern day traffic systems, it is possible that hackers may infiltrate a subset of these vehicles and change their driving parameters. These hacked vehicles, referred to as malicious vehicles, can be arbitrarily interspersed with the other (normal) vehicles and perform a series of coordinated, subtle velocity changes, with the objective of introducing undesirable waves in the traffic system, which can impact the overall vehicle flow and even fragment the road connectivity. Modelling individual vehicles in a traffic system can reveal important information about individual driving behavior and the impact of malicious vehicles, however the computational complexity increases considerably with the number of vehicles considered in the system. On the other hand, macroscopic models, which draw inspiration from Eulerian fluid models, describe the behavior of the traffic by describing lumped or aggregate quantities like density and average velocity, calculated at spatio-temporal intervals. Such models are highly capable of describing collective transport phenomena such as the evolution of congested regions or the velocity of propagation of traffic waves in a computationally efficient manner. In this talk, normal and malicious vehicles are modeled using a two-species macroscopic Partial Differential Equation (PDE) traffic model. The two-species model is analyzed using a combination of analytical and machine learning methods to detect the presence of malicious vehicles in the traffic, as well as quantify their number, distribution and impact on the traffic system. Non-linear PDE simulations highlight the efficacy of these analyses.

Keywords: Cognition modeling, Assistive and Rehabilitation Robotics, Human-Machine and Human-Robot Systems
Abstract: Intelligent Tutoring Systems (ITS) emulate human tutors by closing-the-loop between human learners and tutoring agents. However, in comparison to ITS developed for traditional disciplines (e.g., mathematics or language), developing ITSs for psychomotor skills has its own set of challenges. For example, assessing “correctness” of an answer to a math problem does not directly translate to evaluating correct psychomotor task performance. Key challenges for psychomotor ITS include creating a task knowledge space, personalizing agents to learner characteristics, and maintaining learner motivation. To address these design challenges, we propose a cognitively aware ITS for psychomotor learning with specific consideration of a task where users learn to safely land a quadrotor in a 2D simulator. Cognitive factors such as self-confidence and workload influence learners' self-efficacy and learning outcomes, yet their operationalization in psychomotor ITSs remains limited. In response to this, we train a within flight automation assistance algorithm based on an optimal control policy designed not only to calibrate self-confidence to performance but also calibrate workload to appropriate levels throughout the sequence of landing attempts. We design and leverage a learning stage classifier to quantitatively characterize novice-to-expert performance, bridging qualitative and quantitative learning stage representation. Our policy is trained using reinforcement learning methods with self-confidence, workload, and learning stage Markov Decision Process models. By combining learning stage classification, task performance metrics, and automation assistance, our system generates tailored formative feedback—positive, neutral, or negative—enhancing personalization. This approach addresses critical challenges of psychomotor ITS design, offering a framework for effective ITSs.

Keywords: Cyber physical systems
Abstract: Increased connectivity and automation in modem infrastructures have amplified the vulnerability against cyberattacks that can impair system operation. Rapid detection as well as isolation of these cyberattacks is crucial to ensure reliable and safe operations by dispatching appropriate targeted remedial measures. However, the problem of isolating the attack source based on system model and measurement is inherently ill-posed and often relies on the availability of redundant sensors. In this talk, I will present a data-driven strategy for detection-isolation of cyberattacks on actuation and sensor via the Koopman operator. Specifically, I will cover how the algorithm leverages the changes in Koopman modes and eigenfunctions to detect as well as isolate actuation and sensor cyberattacks, without prior knowledge of the system model or requirement for redundant sensing. The algorithm adopts an online small-data learning strategy that ensures resource-efficient implementation and broad generalizability while addressing the limitations of both traditional model-based and data-driven approaches. The algorithm is applied for attack detection-isolation in compromised electric vehicle charging, where the charging actuation commands and the sensor measurements from the vehicle battery can be corrupted. Finally, I will present case studies with high-fidelity battery simulations using 'PyBaMM' and 'liionpack' under rate-limited measurements and realistic uncertainties to demonstrate the efficacy of the proposed algorithms. I will thereby highlight its potential impact on the safety of real-world energy infrastructure.

Keywords: Human-Machine and Human-Robot Systems, Manufacturing Systems, Modelling, Identification and Signal Processing
Abstract: Please see the document.

Keywords: Machine Learning in modeling, estimation, and control, Robotics, Human-Machine and Human-Robot Systems
Abstract: Ensuring safety in learning-enabled control systems is increasingly vital as neural networks become integral to robotics and other complex dynamical systems. This talk presents recent advances in the formal verification and synthesis of neural control barrier functions (neural CBFs) that guarantee safety in nonlinear systems represented by neural networks. I begin by introducing a verified neural dynamic modeling framework that employs Bernstein polynomial-based over-approximations, which enables real-time safe control and achieves significant speedup over the complete verifier while maintaining safety guarantees. Empowered by our recent neural network verification toolbox, I will then introduce symbolic derivative bound propagation techniques to verify neural CBFs with improved tightness and scalability of formal verification. These methods form the mathematical backbone of two frontier applications. First, in PDE boundary control, I show how neural boundary CBFs ensure constraint satisfaction for systems governed by unknown PDEs via neural operator modeling. Second, I extend these safety concepts to human-AI conversations, modeling dialogue context state transitions in large language models (LLMs) as controllable dynamical systems. Using neural CBFs, we conduct online safety steering that provably prevents LLMs from multi-turn jailbreaking attacks. By unifying methods across physical and symbolic domains, this line of work opens the door to verifiable dynamical safety in a wide range of autonomous systems— from robotic systems to conversational AI agents. The talk emphasizes verifiable forward invariance induced by neural barrier function and the integration of neural dynamics with control theory, demonstrating a scalable path toward trustworthy and safe AI-driven control.

Keywords: Manufacturing Systems, Adaptive and Learning Systems, Control Design
Abstract: Please see my submitted document.

Keywords: Manufacturing Systems, Path Planning and Motion Control, Robotics
Abstract: Please see my attached document.

Keywords: Mechatronic Systems
Abstract: Cooperative control for connected and automated vehicles (CAVs) is a critical component of next-generation intelligent transportation systems, aimed at enhancing traffic safety and management. However, traditional methods like one-dimensional (1D) vehicle platooning are fundamentally limited, as they ignore lateral vehicle interactions and cannot utilize the full capacity of multi-lane roads. This work introduces a novel cooperative control framework, termed "vehicle flocking," designed to overcome these limitations by coordinating the planar motion of multi-agent CAVs as a cohesive flock. A key challenge is applying flocking control within complex, varying, and structured road environments where vehicle behavior is governed by human-defined traffic regulations. To address the challenge, novel vehicle flocking rules are defined that integrate vehicle dynamics, traffic regulations, and permanent road boundaries directly within the three foundational flocking principles: separation, alignment, and cohesion. This integration is achieved through several technical innovations, including a new elliptical lattice-based spacing policy, an artificial potential function to enforce road boundaries, and an artificial flow guidance method to navigate the multi-agent CAVs. The framework's power and versatility are demonstrated in three critical applications: (1) a virtual vehicle-based model for seamless ramp merging, (2) a field-of-view neighbor selection rule for robustness against time delays; and (3) a novel dual-loop MPC structure for systematic parameter tuning. Simulation results confirm this control strategy enables reliable, coordinated motion of multi-agent CAVs, allowing them to maintain stable formations while enhancing overall traffic efficiency.

Keywords: Mechatronic Systems, Robotics, Manufacturing Systems
Abstract: Multi-process, multi-material additive manufacturing has the potential to revolutionize the design and fabrication of complex electronic and electromagnetic systems. In this work, advances in multi-process, multi-material desktop 3D printing for electromechanical actuators and frequency selective surfaces are presented. First, novel 3D-printed axial- and radial-flux motors are introduced. These motors are fabricated with three- or four-axis 3D printers using silver direct ink write (DIW) and plastic fused filament fabrication (FFF). Multi-process printing techniques enable the printing of high efficiency stators and electromechanical actuators and grippers during a single print session within a single additive manufacturing system. This novel technology transforms on-demand manufacturing, enabling seamless in-field component replacement and system upgrades for components and actuators. Next, we explore the fabrication of a doubly-curved, multi-layer frequency selective surface (FSS) sub-reflector for satellite communication with a multi-process desktop 3D printer implementing non-planar printing techniques. With additive manufacturing, it becomes possible to manufacture a complex, non-planar FSS with doubly-curved and parabolic shapes. Here, a triband FSS sub-reflector is designed and printed to expand satellite communication capabilities by separating S and C (2 to 5 GHz) wireless frequency bands from highspeed Ku (17.7 to 20.2 GHz) and Ka (27.5 to 30.0 GHz) frequencies without the need for separate antenna systems. This reduces satellite weight, cost, and complexity. Finally, a pick and place capability integrated within multi-process additive manufacturing is explored, including a demonstration of an axial three-phase electric motor with integrated Hall effect sensor circuitry. Overall, the multi-process additive manufacturing techniques described in the work enable rapid, on-demand prototyping and the manufacturing of mission-critical parts for applications that are beyond the reach of traditional supply chains. These innovations unlock new possibilities for space exploration, resilient infrastructure, and next-generation wireless technologies.

Keywords: Modeling and Control of Biotechnological Systems, Healthcare systems, Modelling, Identification and Signal Processing
Abstract: Acute mental stress arises in many everyday life circumstances in response to perceived threats (i.e., stressors) and negatively impacts the psycho-physiological states of individuals. It has been shown that acute mental stress induces sympathetic arousal, whose frequent occurrence may deteriorate the quality of life. This talk covers our recent studies on transcutaneous median nerve stimulation (tMNS) as an emerging non-invasive modality used for its therapeutic effect on mitigating stress-induced sympathetic arousal. We share our experimental results and developments toward a tMNS-enabled personalized health intervention from the three standpoints of i) sensing, detection, and monitoring, ii) modeling cardiovascular responses to acute stress and tMNS, and iii) closed-loop just-in-time interventions. First, we present a novel acute mental stress detection algorithm based on statistical inference and a novel synthetic multi-modal variable (SMV) that enables real-time and personalized monitoring of cardiovascular responses to acute mental stress and tMNS. The SMV integrates six plausibly explainable physio-markers extracted from three wearable sensing modalities, namely photoplethysmogram (PPG), electrocardiogram (ECG), and seismocardiogram (SCG). The experimental data from healthy individuals suggest that the SMV exhibits superior consistency, sensitivity, and robustness compared to individual physio-markers. We also present an inference-enabled algorithm that leverages the SMV to track and monitor stress-induced cardiovascular arousal. Second, we present a virtual experiment generator (VEG) developed through system identification and variational inference approaches to replicate population-level and subject-specific cardiovascular responses to acute mental stress. VEG enables in silico testing and development. Third, we leverage the VEG and design and evaluate robust control algorithms for just-in-time closed-loop controlled tMNS to effectively mitigate acute stress-induced arousal despite large inter- and intra-individual variability. Overall, this talk presents the system-theoretic challenges and opportunities of developing closed-loop tMNS interventions for mitigating stress-induced sympathetic arousal.

Keywords: Modeling and Validation, Manufacturing Systems, Control Design
Abstract: Please see the document.

Keywords: Modelling, Identification and Signal Processing, Optimal Control, Control Design
Abstract: Proton Exchange Membrane (PEM) fuel cells are a leading clean energy technology for transportation and stationary applications due to their high efficiency and low emissions. Their performance, however, is highly dependent on maintaining optimal membrane hydration, which is governed by internal humidity conditions. This study develops control-oriented models and strategies to manage humidity within PEM fuel cells. A lumped-parameter physics-based model was developed in MATLAB/Simulink to predict internal humidity dynamics. While grounded in physical laws, such models can struggle with accuracy due to assumptions and the complexity of PEM fuel cells. To address this, data driven modeling techniques were explored. In particular, the Koopman operator, a linear but infinite-dimensional operator capable of capturing nonlinear dynamics, was used. Finite dimensional approximations were constructed using time-delay embeddings and radial basis functions. A NARX neural network was also investigated and compared to the Koopman model. Among the models studied, the Koopman model with time-delay embeddings outperformed both the physics-based and NARX models, showing strong potential for capturing nonlinear water management behavior in PEM fuel cells. This Koopman model was then employed in a model predictive control framework (KMPC) as the predictive model, while the physics-based model served as the plant to optimize anode relative humidity and performance in an open-cathode PEM fuel cell stack. The performance of KMPC was compared to a baseline proportional-integral (PI) controller. While both achieved similar reference tracking, KMPC adjusted control effort based on operating conditions, improving efficiency. Additionally, KMPC’s linearity enables the use of efficient linear MPC solvers, making it well-suited for real-time control.

Keywords: Motion and Vibration Control, Manufacturing Systems
Abstract: Electrohydrodynamic (EHD) printing has become a promising and cost-effective technique for producing high-resolution and large-scale features. One widely recognized obstacle in EHD printing is nozzle clogging due to solvent evaporation or ink polymerization. Moreover, printing highly viscous materials often requires pressure or other external force to assist the ink flow during the printing, which increases the complexity of process control and the required energy. In this work, we developed a novel ultrasonic vibration-assisted EHD printhead and associated process to effectively eliminate the nozzle clogging for the printing of high-viscosity and high-evaporation-rate inks. A series of experimental tests were conducted to characterize the printhead design and process parameters (i.e., vibration frequency, vibration amplitude, and printing voltage). The results demonstrated that superimposing ultrasonic vibration on the EHD printing nozzle can effectively enhance current EHD printing capabilities, such as reducing required pressure, eliminating nozzle clogging, and providing stable and continuous printing for high viscosity and high solvent evaporation rate material. With the optimal parameters, a filament with a diameter of around 1μm can be continuously printed, and we successfully applied this developed ultrasonic-assisted EHD process to print high-resolution 2D patterns.

Keywords: Optimal Control, Control Applications
Abstract: Manufacturing industries face increasing pressure to improve sustainability while maintaining production efficiency and meeting customer demands. As renewable energy integration accelerates and electricity markets shift toward dynamic pricing structures, manufacturing systems must evolve beyond traditional scheduling paradigms to become grid responsive. However, at scale, these energy-aware manufacturing control schemes must be responsive to 1) time-varying electricity prices and renewable energy availability, 2) complex networked production flows and inventory constraints, and 3) customer order requirements and production targets. This talk will present recent results on hierarchical model predictive control methods that enable manufacturing systems to be both customer-responsive and energy-responsive. We will explore how networked manufacturing system models can capture complex material flows and energy consumption patterns across interconnected machines and buffers. The presentation will demonstrate how model predictive control frameworks can dynamically optimize production schedules in response to real-time pricing while maintaining production commitments. Finally, we will examine bi-level optimization approaches that jointly optimize product pricing and production scheduling, enabling manufacturers to shape demand while leveraging renewable energy availability. Through case studies involving battery manufacturing systems, we will show how these integrated approaches can achieve significant energy cost reductions while maintaining profitability and production targets. The talk will highlight the potential for manufacturing systems to become active participants in demand-side energy management, contributing to grid stability while enhancing their own operational efficiency and sustainability.

Keywords: Power and Energy Systems, Transportation Systems
Abstract: This presentation surveys the literature on barrier function-based safety control and identifies a critical research gap, namely, the need for safety control algorithms that degrade gracefully in the presence of system failures and anomalies. The presentation proposes a novel control design paradigm that embeds the notion of graceful degradation within the control barrier function theory. Intuitively, the idea is to switch from one safety control mode to another in response to changes in the system's environment and/or damage/hazard state. Mathematically, we achieve this by modifying the design of the barrier functions and constraints to create a multi-layered definition of safety. We illustrate this approach for two different application problems, namely, preventing thermal runaway propagation in lithium-ion batteries and ensuring graceful collision avoidance in adaptive cruise control applications.

Keywords: Robotics
Abstract: Autonomous systems are rapidly reshaping domains like transportation, manufacturing, and everyday human life. Yet, their widespread deployment hinges on a critical question: Can we ensure these systems consistently behave as intended—never in ways we cannot predict or control? This talk presents a cohesive research agenda that tackles this challenge through three interconnected thrusts: (i) expressive task specification, (ii) correct-by-construction planning and control, and (iii) formal robustness certification.

I begin by introducing hierarchical temporal logic specifications that express rich combinations of safety and liveness goals while remaining intuitive for engineers to write—and even translatable from natural language via large language models. These representations dramatically reduce specification time and scale to complex multi-robot coordination tasks.

Next, I highlight a suite of planning, control, and decision-making algorithms that rigorously fulfill such specifications. This includes the first abstraction-free motion planner with both probabilistic completeness and asymptotic optimality, scalable task allocation and motion planning for multi-robot collaboration, and methods that leverage previously solved problems through plan reuse.

Finally, I demonstrate how these guarantees extend into modern learning-based components. In collaboration with Boeing, I present the first formal certification of 6-DoF pose estimation pipelines built on keypoint detection, even under challenging visual conditions such as occlusion and more complex semantic perturbations.

Combined with the principles of specification, synthesis, and verification, these contributions lay the foundation for autonomous systems that are both capable and backed by formal guarantees of reliability.

Keywords: Robotics
Abstract: Robotics foundation models pre-trained on internet-scale data have started to revolutionize how robots understand the complex world, interpret human feedback, and plan actions. The promise of using these models within robotics is the ability to generalize robot behaviors and push robot deployment into increasingly unstructured or novel environments. However, despite the remarkable progress, it is precisely this integration of foundation models that introduces new safety and alignment challenges in robotics. Robots are safety-critical systems, wherein a foundation model’s single erroneous visual or language interpretation, misaligned behavior generation, or high inference latency can lead to catastrophic consequences. In this talk, I will introduce our recent efforts to generalize model-based control principles to tackle the challenges that have emerged throughout the lifecycle of robotics foundation models, ranging from pre-training to post-training and deployment. On the pre-training side, I will introduce our work on identifying informative “system-level” model failures for cost-effective and targeted model training. On the post-training side, I will introduce our efforts to bring the success of preference alignment, widely adopted in non-embodied foundation models (e.g., large language models), to embodied contexts such as robot manipulation and autonomous driving, enabling robots to align their behavior with human preferences with minimal human feedback. Finally, on the deployment side, I will share our work on test-time model monitoring and adaptation for safe deployment of robotics foundation models in the novel environments.

Keywords: Robotics, Mechatronic Systems, Soft Robotics
Abstract: This talk presents a structure-driven approach to robotic grasping and manipulation, in which key control and modeling challenges are simplified through mechanical design innovation. We introduce mechanical closure, extending classical closure definitions via adaptive structures such as multi-segment soft actuators and origami-based grippers. These enable delicate and robust grasps in complex scenarios—from capturing jellyfish underwater to handling raw egg yolks. To simplify contact uncertainty, we propose adaptive variable stiffness phalanges, allowing reliable interaction across diverse geometries and friction conditions. For in-hand manipulation, the DexCo Hand, a soft–rigid hybrid system with soft hydraulic actuation, reduces control burden through intrinsic compliance. Building on this foundation, we further incorporate selective learning-based frameworks that transfer human-inspired skills—such as the autonomous in-hand manipulation required to in-hand counting and sorting pills. These systems demonstrate how embedding intelligence into mechanical structure, with minimal sensing and model dependence, leads to robust, dexterous, and scalable manipulation. My approach brings these three elements—structure, control, and AI—into synergy, instead of opposition.

Keywords: Robotics, Path Planning and Motion Control, Optimal Control
Abstract: Set-based methods are widely used in control analysis and design. Example applications in robotics include verifying the safety of system trajectories and defining constraints to which a controller must adhere. In the context of reachability analysis, zonotopic sets (i.e., zonotopes and their generalizations) have become popular in recent years because they have closed-form expressions for many important set operations and favorable complexity growth when compared to more traditional set representations. A downside to these sets is that analysis generally relies on numerical optimization.

In this talk, the coupled problems of designing zonotopic sets to be efficiently analyzed and designing optimization algorithms to analyze these sets are considered. Further, reachability analysis using zonotopic sets is conceptualized as a general methodology for building structured optimization problems for dynamic systems. Sparsity-promoting reachability calculations are presented that have lower memory complexity when compared to typical methods. Optimization problems built using these calculations are efficiently solved using tailored approaches based on the alternating direction method of multipliers (ADMM). Optimization times are shown to be faster than state-of-the-art solvers and problem formulations. This reachability analysis and structured optimization workflow can be implemented online to facilitate real-time control, estimation, and safety verification. The proposed framework is applied to both convex (using constrained zonotopes) and non-convex (using hybrid zonotopes) problems, such as those involving hybrid systems and disjoint constraint sets. New methods are experimentally validated in application to path and motion planning for ground robots.

Keywords: Soft Robotics, Mechatronic Systems, Sensors and Actuators
Abstract: Soft robots are increasingly utilized for exploring unstructured environments and delicately handling objects, primarily due to their inherent compliance and safe interaction with humans. To enable soft robots with embodied perception, it is essential to acquire accurate proprioceptive feedback and external environmental perception (exteroception). However, unlike traditional rigid robots equipped with precise rigid sensors like encoders, soft robots face significant challenges in proprioception due to their infinite degrees of freedom. Additionally, it is crucial for soft robots to estimate external forces without exterior rigid sensors when interacting with the environment in applications such as minimally invasive surgery. This talk presents an integrated approach to simultaneous continuous shape reconstruction and external force estimation for soft robots using a novel distributed inductive curvature sensor. This sensor captures continuous actuator shapes in real-time through electromagnetic induction, enabling high accuracy and practical scalability. In addition, an enhanced analytical model based on the Euler–Bernoulli curved beam theory is developed to predict the shape under pneumatic actuation and external forces. External forces are estimated through a model-based optimization approach based on the measured shape. Furthermore, this talk will also introduce a learning-based modeling framework leveraging a Euler spiral-inspired shape representation for soft robots. Using this representation, we develop neural network-based forward and inverse models that accurately predict the actuator's shape and estimate external forces. In summary, this talk presents advances in embedded sensing and learning-based modeling that significantly enhance embodied perception—integrating proprioception and exteroception—and pave the way for more intelligent, adaptive, and autonomous soft robotic systems.

Keywords: Soft Robotics, Modeling and Validation, Robotics
Abstract: Soft continuum robots offer benefits compared to conventional rigid robots for applications in medical settings, grasping/handling objects, and navigating confined and unstructured environments due to their inherent compliance. A highly sought after goal is to replicate the dexterity and manipulability of biological structures, such as elephant trunks. However, the continuous form and underactuation of soft robots pose challenges in modeling and control and is an active research area. In this talk, I will describe a modeling framework based on a constrained Lagrangian approach in non-minimal coordinates that can accommodate holonomic and non-holonomic constraints and capture the dynamic characteristics of soft continuum robots. The simplicity and generalizability of this approach is unique and is currently not replicated in the soft robot modeling literature. To demonstrate the model's efficacy, we designed and validated two highly dynamic soft robots. The first is biologically-inspired and mimics the motion of a Pacific-lamprey fish. It is able to scale a burlap-lined vertical wall up to 10 cm/sec (0.47 body lengths per second, blps). The second is a wheeled robot that is able to locomote over a planar surface up to 22 cm/sec (1.38 blps), reverse, and perform circular motion. Both these robots are modeled with this framework and their configuration and velocities are characterized over a broad frequency range (0.5 to 5 Hz) demonstrating a close match with simulation. We are currently extending this open-loop validation for closed-loop, model-based control of a 2-DOF soft actuator. The soft actuator is designed to enable grasping and facilitate environmental interaction by increasing curvature along its backbone to the tip. Preliminary simulation results indicate robustness to parametric uncertainty and low tracking error for line and circle following tasks at relatively high frequency. The implementation of real-time robust control for this class of highly dynamic underactuated soft actuators will significantly improve soft robot functionality.

Keywords: Agricultural Systems, Unmanned Ground and Aerial Vehicles, Intelligent Autonomous Vehicles
Abstract: Biological control is a means of removing problem insects, plants, and pathogens with human assisted solutions through the use of natural enemies. In Southwestern Texas and Mexico, one such problem insect which impacts both local and regional economies is the invasive cactus moth (Cactoblastis cactorum). The cactus moth has been a major concern of the United States Department of Agriculture (USDA), and Mexico’s Secretariat of Agriculture and Rural Development, as it threatens the Mexico’s 74% share in the world production of prickly pear cactus and prickly pear derivative products. Computer vision is a growing area of artificial intelligence which offers an effective way to detect and diagnose harmful insects and pathogens with minimally invasive and cost-effective means. This paper proposes an indoor computer-vision-based localization method for the effective deployment of biological control using a transfer learning approach to the YOLOv8 architecture. YOLOv8 proved to be an effective model for the proof-of-concept health monitoring segmentation attaining an accuracy of 55% and a tested object detection deployment inference time of 300ms. Overall, this system demonstrates an efficient and effective approach to image-based localization ready for outdoor testing and deployment of biological control within small to medium-sized agricultural environments.

Keywords: Biomechanical Systems, Cyber physical systems, Modelling, Identification and Signal Processing
Abstract: Curriculum planning for many educators and educational researchers takes into account the use of the classroom environment. This is true across most age groups, but becomes even more important for early childhood learning for which the design of the space, the activities associated with a particular location, and the learning objectives are tightly coupled. Existing approaches to understand classroom usage are based on human-based observational methods and qualitative assessments. However, these methods may result in misallocation of the spatial usage when multiple students are engaged in different areas within the environment and the human observer may not be present. To provide a more quantitative approach, this work investigates the use of non-invasive wearable sensors and Markov chains to analyze proximity data collected for 8 days over a span of 4 months. Probabilistic maps representing the transition sequences between different location and task states for 7 users are presented. Variations between users and from day-to-day interactions are discussed. The Markov chains derived from these data illustrate that students preferred the snack area when the teacher was not engaged within the decision-making process, with a shift to drama time or circle time when the teacher was engaged with the students.

Keywords: Modeling and Validation, Human-Machine and Human-Robot Systems, Biomechanical Systems
Abstract: Synthetic data offers a versatile solution for generating realistic datasets. In this study, we introduce a methodology specifically focused on generating synthetic inertial acceleration data that effectively captures the dynamics of human arm gestures. Classical human biomechanical models for human motion—such as minimum jerk, minimum torque-change, and iterative Linear Quadratic Regulator (iLQR) are adapted and used to systematically decompose complex gestures into fundamental point-to-point movements. The models incorporate realistic sensor noise models to enhance the authenticity of the synthetic datasets, and experimental validation demonstrates that the synthetic trajectories generated closely align with actual inertial data recorded during predefined arm gestures.

Keywords: Modelling and Control of Biomedical Systems, Control Applications, Assistive and Rehabilitation Robotics
Abstract: Muscle fatigue presents a major challenge in Functional Electrical Stimulation (FES)-based rehabilitation, where early fatigue onset can reduce therapeutic efficacy and session duration. While electromyography (EMG) and ultrasound have been used to monitor fatigue, these methods can be costly, complex, or prone to interference during stimulation. This study explores the use of mechanomyography (MMG) as a low-cost, lightweight alternative for fatigue monitoring during sustained isometric contractions of the biceps brachii. MMG signals were acquired via a wearable accelerometer, and time- and frequency-domain features—root mean square (RMS) amplitude and mean power frequency (MPF)—were extracted. A fatigue threshold was defined based on consistent increases in RMS and decreases in MPF relative to baseline, and reliably identified fatigue onset between 60–80% of endurance time across participants and contraction levels. The proposed MMG-based threshold demonstrated consistent performance, highlighting its potential for real-time fatigue monitoring in adaptive FES systems and broader rehabilitation applications.

Keywords: Modelling and Control of Biomedical Systems, Healthcare systems, Nonlinear Control Systems
Abstract: Pulmonary fibrin accumulation is caused by aberrant cascade interactions that promote fibrin protein deposits in the lung. Fibrin buildup stiffens the lungs, decreases gas exchange, and impairs lung function. Unregulated fibrin accumulation can devolve into irreversible scarring, called fibrosis, and ultimately lung failure. Diseases with pulmonary fibrin accumulation such as acute respiratory distress syndrome (ARDS) have no known cure. Supportive care that is open loop and non-personalized is the only treatment option. Here, we propose closing the loop for ARDS with a feedback treatment regimen of a novel protein therapeutic, activated protein C (APC). We show the observability, controllability, and feedback linearizability of a single input, single output control-affine nonlinear system that captures the simplified dynamics of pulmonary fibrin accumulation during virally-induced ARDS. We then apply our developed input-state feedback linearizing APC controller to our simplified system, and demonstrate the return of a diseased patient to a healthy reference.

Keywords: Modelling and Control of Biomedical Systems, Optimal Control, Nonlinear Control Systems
Abstract: This paper introduces a novel autonomous fluid resuscitation algorithm that maintains hemodynamic stability despite sparse, noisy clinical measurements. We develop a robust nonlinear state‐space modeling (RNSSM) framework, trained via a variational autoencoder, to predict mean arterial pressure (MAP) in response to fluid infusion during hemorrhage. This model is paired with a novel radial basis function (RBF) optimal control strategy that employs predictive optimization with function approximation to compute subject-specific fluid administration rates. The RNSSM’s accuracy was validated on real animal data, and simulation results demonstrated that the RBF controller outperformed state-of-the-art resuscitation algorithms, addressing key limitations of existing methods in critical care.

Keywords: Adaptive and Learning Systems, Nonlinear Control Systems, Assistive and Rehabilitation Robotics
Abstract: Spasticity, muscle weakness, and contracture in the musculotendon complex of the ankle joint are debilitating factors after a stroke or neurological injury. Motorized rehabilitation techniques involving dynamic stretching of the ankle plantarflexors in people with spasticity can improve passive stiffness, maximum voluntary contraction, and range of motion. However, clinical work has leveraged heuristic open-loop torques or PID control of ankle devices, which may lead to unsafe human-machine interactions and lack of customization to evoke desired neuromuscular responses. This paper develops a continuous closed-loop adaptive controller leveraging a desired compensation adaptive law-like approach (using desired trajectories in the regressor) and a high-order tuner to track kinematic trajectories designed to evoke ankle plantarflexor stretches using a motorized ankle device. A high-order tuner strategy is used to improve transient performance and to generate smoother parameter estimates when compared to standard gradient-based adaptive controllers. A Lyapunov-based stability analysis guarantees globally uniformly ultimately bounded (GUUB) kinematic tracking.

Keywords: Adaptive and Learning Systems, Multi-agent and Networked Systems, Control Design
Abstract: This manuscript considers the problem of ensuring stability and safety during formation control with distributed multi-agent systems in the presence of parametric uncertainty in the dynamics and limited communication. We propose an integrative approach that combines Adaptive Control, Control Barrier Functions (CBFs), and connected graphs. The main elements employed in the integrative approach are an adaptive control design that ensures stability, a CBF-based safety filter that generates safe commands based on a reference model dynamics, and a reference model that ensures formation control with multi-agent systems when no uncertainties are present. The overall control design is shown to lead to a closed-loop adaptive system that is stable, avoids unsafe regions, and converges to a desired formation of the multi-agents. Numerical examples are provided to support the theoretical derivations.

Keywords: Adaptive and Learning Systems, Nonlinear Control Systems
Abstract: This paper develops an input-output feedback linearization-based adaptive controller to stabilize and regulate a dual-rotor rotational system (DRRS), whose inertial properties as well as the geometric configuration of rotors are unknown. First, the equations of motion governing the dynamics of DRRS are derived using the Newton-Euler approach. Next, an input-output feedback linearization technique is used to linearize the dynamics from the rotor speeds to the angular position of the system. A finite-time convergent estimator, based on the portion of the DRRS dynamics, is used to update the required parameters in the controller. Finally, the proposed controller is validated in both step and harmonic command-following problems, and the robustness of the controller to the system's parameters is demonstrated.

Keywords: Adaptive and Learning Systems, Modelling, Identification and Signal Processing
Abstract: In recent years, adaptive identification methods that can achieve the true value convergence of parameters without requiring persistent excitation (PE) have been widely studied, and concurrent learning has been intensively studied. However, the parameter convergence rate is limited for the gradient-based method owing to small parameter update gain, and even the introduction of forgetting factors does not work sufficiently. To address this problem, this study proposes a novel discrete-time recursive least squares method under finite excitation (FE) conditions using two forgetting factors (inner and outer) and an augmented regressor matrix comprising a sum of regressor vectors. The proposed method ensures the PE condition of the augmented regressor matrix under FE conditions of the regressor vector and allows the properly design of the forgetting factor without estimator windup and/or destabilization of the system. Numerical simulations demonstrate its effectiveness by comparing it with several conventional methods.

Keywords: Adaptive and Learning Systems, Nonlinear Control Systems, Control Design
Abstract: This paper introduces a distributed data-driven attack-resilient consensus problem under both false data injection (FDI) and denial-of-service (DoS) attacks and proposes a data-driven consensus control framework, consisting of a group of comprehensive attack-resilient data-driven observers. The proposed group of observers is designed to estimate FDI attacks, external disturbances, and lumped disturbances, combined with a DoS attack compensation mechanism. A rigorous stability analysis of the approach is provided to ensure the boundedness of the distributed neighborhood estimation consensus error. The effectiveness of the approach is validated through numerical examples involving both leaderless consensus and leader-follower consensus, demonstrating significantly improved resilient performance compared to existing data-driven control approaches.

Keywords: Linear Control Systems, Estimation, Control Design
Abstract: This paper introduces a new method to transform any controllable and observable linear time-invariant system into a strictly positive real system using a proportional-integral (PI) observer. Strictly positive real systems are crucial in many control applications, including robust stabilization and passive control design. This new method has the benefit of canceling the noise from the source signal while demonstrating the ability to transform both stable and unstable systems into strictly positive real systems with a PI observer.

Keywords: Soft Robotics, Machine Learning in modeling, estimation, and control, Estimation
Abstract: Soft robots, distinguished by their inherent compliance and continuum structures, present unique modeling challenges, especially when subjected to significant external loads such as gravity and payloads. In this study, we introduce an innovative data-driven modeling framework leveraging Euler spiral-inspired shape representations to accurately describe the complex shapes of soft continuum actuators. Based on this representation, we develop neural network-based forward and inverse models to effectively capture the nonlinear behavior of a fiber-reinforced pneumatic bending actuator. Our forward model accurately predicts the actuator’s deformation given inputs of pressure and payload, while the inverse model reliably estimates payloads from observed actuator shapes and known pressure inputs. Comprehensive experimental validation demonstrates the effectiveness and accuracy of our proposed approach. Notably, the augmented Euler spiral-based forward model achieves remarkably low average positional prediction errors of just 3.38%, 2.19%, and 1.93% of the actuator length at the one-third, two-thirds, and tip positions, respectively. Furthermore, the inverse model demonstrates exceptional precision, estimating payloads with an average error as low as 0.72% across the tested range. These results underscore the potential of our method to significantly enhance the accuracy and predictive capabilities of modeling frameworks for soft robotic systems.

Keywords: Machine Learning in modeling, estimation, and control, Cognition modeling, Assistive and Rehabilitation Robotics
Abstract: Dynamic haptic feedback policies mediated through human-robot interaction (HRI) can accelerate hand motor skill acquisition. Yet, online methods struggle to handle the numerous degrees of freedom and complex motor planning dynamics of the hand. Iterative methods can simplify complex policy optimization problems but require significant training data from naive participants. This work introduces NeuroSiGHT, a Neuromechanical Simulation for Generalized Human Trials, combining emerging neuromotor and biomechanical models to simulate multisensory perception, dynamics of HRI, and motor cognition of healthy human participants in motor learning paradigms. NeuroSiGHT offers a computationally efficient platform to warm start iterative haptic policies for efficient deployment.

Keywords: Machine Learning in modeling, estimation, and control, Robotics
Abstract: This paper presents a novel bio-inspired robotic jellyfish with an artificial swim bladder. To control buoyancy, reversible proton exchange membrane (PEM) electrolysis is utilized for the artificial swim bladder, while a direct current (DC) motor-driven artificial bell serves as the primary propulsion system. A physics-based model of the jellyfish robot was developed, and real-time experiments were conducted in an aquarium environment to achieve closed-loop depth control. The identification of the system was performed using the experimental results to estimate the transfer function of the system, which is then applied in the design of predictive model control (MPC). A proportional-integral-derivative (PID) control strategy is implemented for the swim bladder, and a reinforcement learning (RL)-enhanced MPC is designed for the propulsion system. The effectiveness of combining PID control with RL-driven MPC is demonstrated through simulation case studies, which utilize the physics-based model and the results of the identification of the system. These studies show high performance and robustness, even in the presence of disturbances.

Keywords: Machine Learning in modeling, estimation, and control, Soft Robotics
Abstract: Precise kinematic modeling is critical in calibration and controller design for soft robots, yet remains a challenging issue due to their highly nonlinear and complex behaviors. To tackle the issue, numerous data-driven machine learning approaches have been proposed for modeling nonlinear dynamics. However, these models suffer from prediction uncertainty that can negatively affect modeling accuracy, and uncertainty quantification for kinematic modeling in soft robots is underexplored. In this work, using limited simulation and real-world data, we first investigate multiple linear and nonlinear machine learning models commonly used for kinematic modeling of soft robots. The results reveal that nonlinear ensemble methods exhibit the most robust generalization performance. We then develop a conformal kinematic modeling framework for soft robots by utilizing split conformal prediction to quantify predictive position uncertainty, ensuring distribution-free prediction intervals with a theoretical guarantee.

Keywords: Nonlinear Control Systems, Assistive and Rehabilitation Robotics, Uncertain Systems and Robust Control
Abstract: Hamstring spasticity in people with spinal cord injury (SCI) causes severe lower limb dysfunction. Static stretching is a primary treatment for spasticity. Active-Isolated stretching (AIS) is an effective strategy in rehabilitation and exercise science for reducing tension in overactive (i.e., spastic) muscles and improving range of motion (ROM). AIS is a more efficient lengthening technique due to its ability to bypass the myotactic (i.e., shortening) reflex that static stretching is subject to. Motivated to expand the scope of care for people with SCI, this paper develops novel closed-loop nonlinear controllers to facilitate AIS of the hamstrings. A robust electric motor kinematic controller is developed to move the leg from its initial supine position to the target end ROM by maintaining the leg within a kinematic range. When the leg reaches the end ROM, functional electrical stimulation (FES) is applied to the quadriceps to reciprocally inhibit (i.e., lengthen) the hamstrings, per AIS protocol. A custom FES controller tracks a desired active muscle torque using feedback from a load cell. An experiment was conducted to demonstrate safe control of AIS. A Lyapunov-based stability analysis is developed to guarantee exponential tracking of the motor and FES control loops. An output strictly passive condition is used to examine the influence of the FES-induced torque on the motor closed-loop error system when the leg is at the end ROM and both controllers are active.

Keywords: Soft Robotics, Control Applications, Modeling and Validation
Abstract: Over the past decades, various types of pneumatically driven soft manipulators have been designed, modeled and controlled to achieve different goals. The soft manipulator presented in this paper mainly consists of bellows segments based on pneumatic networks that enable spherical movements while being constrained in length along their central axes. This elongation constraint permits larger curvature angles and facilitates the use of alternative sensors, such as spatial bending sensors instead of rope-based sensors. Consequently, the individual actuator kinematics of each bellows segment are not measurable and need to be reconstructed. Therefore, a reconstruction method independent of the actuator's elongation using a linearly scaled zenithal projection and a barycentric coordinate system is presented in this paper. Hence, the kinematic configuration and actuator models of the components of the soft manipulator can be fully defined. The models are validated experimentally. The use of a kinematic PI controller enables to fulfill a pick-and-place task showing the capability to manipulate objects in all six spatial degrees of freedom with the aid of a pneumatic network-based turntable.

Keywords: Estimation
Abstract: Popular Bayes filters typically rely on linearization techniques such as Taylor series expansion and stochastic linear regression to use the structure of standard Kalman filter. These techniques may introduce large estimation errors in nonlinear and non-Gaussian systems. This paper overviews a recent breakthrough in filtering algorithm design called textit{N}atural Grtextit{a}dient Gaussiatextit{n} Apprtextit{o}ximation (NANO) filter and compare its performance over a large class of nonlinear filters. The NANO filter interprets Bayesian filtering as solutions to two distinct optimization problems, which allows to define optimal Gaussian approximation and derive its corresponding extremum conditions. The algorithm design still follows the two-step structure of Bayes filters. In the prediction step, NANO filter calculates the first two moments of the prior distribution, and this process is equivalent to a moment-matching filter. In the update step, natural gradient descent is employed to directly minimize the objective of the update step, thereby avoiding errors caused by model linearization. Comparative tests are conducted on four classic systems, including the damped linear oscillator, sequence forecasting, modified growth model, and robot localization, under Gaussian, Laplace, and Beta noise to evaluate the NANO filter's capability in handling nonlinearity. Additionally, we validate the NANO filter's robustness to data outliers using a satellite attitude estimation example. It is observed that the NANO filter outperforms popular Kalman filters family such as extended Kalman filter (EKF), unscented Kalman filter (UKF), iterated extended Kalman filter (IEKF) and posterior linearization filter (PLF), while having similar computational burden.

Keywords: Estimation, Unmanned Ground and Aerial Vehicles, Robotics
Abstract: This paper examines the effect of GNSS spoofing on GNSS-IMU fusion-based position, velocity and attitude estimates. First, it proposes and tunes an estimator considering also IMU sensor biases and real flight data of a multicopter. Then it feeds the estimator with IMU data from a fixed flight trajectory and GNSS data from different trajectories with increasing divergence from the fixed one simulating a perfect spoofing scenario. Detailed examination of the estimates shows that spoofing has non negligible effect on velocity and especially attitude estimates. Thus any spoofing detection algorithm can not be based on attitude estimates which utilize GNSS data.

Keywords: Estimation, Electromechanical systems
Abstract: In situ detection and avoidance of lithium plating during battery charging is a key element to enable fast charging of Li-ion batteries. In this paper, a methodology is proposed to identify occurrence of lithium plating using the cell voltage during charging. The proposed method applies short interruptions, such as 5 seconds, to the constant current charging profile and monitors the cell behavior during the current cutoff period. It is found that, the voltage droop during the current cutoff behaves differently at high charging rates, when reversible plating happens on the anode, compared to low current charging conditions. The voltage droop is used to estimate a low frequency impedance for the cell which is found as a reliable indicator for the plating detection. Validated with anode potential measured using a reference electrode, the results indicate that lithium plating is successfully detected for two different chemistries using the proposed active diagnostics algorithm. In addition, it is found that the estimated impedance behavior is correlated to quantity and severity of the lithium plating observed in the cell.

Keywords: Estimation, Modeling and Validation, Uncertain Systems and Robust Control
Abstract: Human pose estimation is crucial in computer vision for applications like action recognition, virtual reality, and motion analysis. This paper proposes a 3D pose estimation approach integrating monocular or multi-camera input with a human kinematic model and a Kalman Filter (KF) framework for improved accuracy. Refining keypoints using kinematic constraints and Kalman filtering, the method addresses occlusion and motion artifacts such as blur. Experiments on benchmark datasets and Mocap data show that proposed approach enhances robustness in challenging conditions, outperforming vision-only models. This scalable solution benefits applications in biomechanics, human-computer interaction, and sports analytics.

Keywords: Estimation, Modelling, Identification and Signal Processing
Abstract: Permanent Magnet Synchronous Motor (PMSM) serves as a critical component in numerous industrial applications. Robust rotor speed and position estimates are essential for achieving precise motor control. Conventional approaches typically employ full-order nonlinear filters for state estimation. While since the winding currents are directly measurable, deploying a full-order observer to estimate all states becomes redundant. Furthermore, inherent system uncertainties - including modeling inaccuracies, temperature variations, and aging effects - can significantly compromise estimation accuracy. To address these challenges, a novel nonlinear reduced-order H_{infty} filter is introduced to estimate the speed and position of a two-phase PMSM with biased winding resistance. Simulation results show that the proposed filter brings robust estimation results compared to both full- and reduced-order EKFs if there exists model uncertainty. In addition, it can also be transformed to a reduced-order EKF for applications with accurate system dynamics as well.

Keywords: Estimation, Nonlinear Control Systems, Control Design
Abstract: An asymptotic state variable observer is proposed for a bilinear dynamical system with the distinguishing feature that the error dynamics are globally asymptotically stable and independent of the applied input. Only the speed of convergence of the error dynamics may be input dependent. The approach is to apply classical Lyapunov stability theory using a convex optimization algorithm and linear matrix inequality (LMI) tools to in effect isolate the stability property of the error dynamics from the input. The LMIs are used to turn the nonconvex problem into a convex problem. The method is demonstrated on an induction motor drive.

Keywords: Aerospace, Nonlinear Control Systems, Control Design
Abstract: Missions involving rapid and large-angle attitude maneuvers have been conceived for astronomical and Earth observation satellites in recent years. In light of the necessity for an actuator capable of generating large torque, it is imperative to consider the characteristics of an actuator when designing a control system. Actuators capable of generating large torques include the reaction control system (RCS). RCS provides an on/off input by using the reaction force of fuel injection from the thrusters, it can generate a large torque. This paper considers large angle spacecraft attitude maneuvers by RCS. To this end, a design model for controller design considering the characteristics of RCS is derived based on the relative motion equation of the spacecraft. Next, A nonlinear tracking controller based on sliding mode control is proposed to guarantee boundedness of the tracking error in the presence of system uncertainties and external disturbances. Then, we propose a method to appropriately change RCS injection threshold according to spacecraft attitude by solving an optimization problem. Finally, the effectiveness of the proposed control method is verified by numerical simulations.

Keywords: Aerospace, Unmanned Ground and Aerial Vehicles, Machine Learning in modeling, estimation, and control
Abstract: Helicopter aerial refueling is a particularly challenging maneuver because of the complex aerodynamic interaction between the helicopter, the hose-drogue, and the tanker. To address this, this paper presents a control design and analysis framework for autonomous helicopter aerial refueling. The control architecture is based on the standard cascaded inner-outer loop helicopter control based on dynamic inversion. The outer-loop dynamic-inversion based control is augmented by a reinforcement learning (RL) controller that corrects the outer-loop commands to account for the unpredictable drogue motion. The RL corrective input along with inner loop tracking error result in imperfect dynamic inversion in the outer-loop, leading to a nonlinear residual term in the outer loop dynamics. Hence, we derive analytical stability and performance bounds of the proposed controller in the presence of bounded drogue uncertainty, RL control actions, and imperfect inner loop tracking. Simulations in a high-fidelity environment with a full-scale helicopter and a high-fidelity drogue model validate the performance of the proposed method. The RL agent is trained on a reduced-order helicopter model, and then transferred directly to a full-scale helicopter simulation platform without any retraining. The docking simulations conducted in the full-scale helicopter model reveal that we can obtain a mean docking error of 0.31m with a standard deviation of 0.05m with the pure model-based controller in 100 simulation runs, while the proposed controller reduces the mean docking error to 0.17m with a standard deviation of 0.07m, demonstrating an improvement of 45%.

Keywords: Control Design, Uncertain Systems and Robust Control, Aerospace
Abstract: The optimization of flight control parameters is caught between two conflicting objectives: Striving for the best performance while achieving sufficient robustness. The problem of robustness is concerned with the variations of system and environmental parameters. In that sense, a controller must still provide satisfactory performance under off-nominal conditions. To address this issue, different approaches have been proposed that either include multiple parameter sets in the optimization or use different metrics to force the optimization into a robust solution. Nevertheless, guarantees on the performance under any parameter variations cannot be made within these approaches. For this reason, this paper proposes an algorithm that includes worst case searches in the controller optimization such that not only nominal soft and hard constraints are satisfied but also worst case constraints under any expected parameter variation.

Keywords: Electromechanical systems, Mechatronic Systems, Unmanned Ground and Aerial Vehicles
Abstract: Oscillatory or vibratory motion generation is energetically demanding. It consists of energy dissipation in the damping or resistive components, and the energy required to accelerate and decelerate the oscillator. While the former cannot be avoided, the latter can be reduced or eliminated using resonance. Here we present a novel way of achieving resonance using an inductor connected in series with a direct drive DC motor. Specifically, resonance is achieved between the kinetic energy of the oscillator and the electrical energy of the inductor, resulting in reduction in energy expenditure from power source as well as reduction in voltage amplitude required to achieve a given oscillatory motion. While capacitors are also reactive electrical components, we show that a capacitor in series with a DC motor always results in an overdamped system with limited resonant amplification. Experiments verify the resonance phenomenon and energetic benefits of using an inductor.

Keywords: Estimation, Unmanned Ground and Aerial Vehicles, Sensors and Actuators
Abstract: Airborne Wind Energy (AWE) exploits kites for high-altitude power generation but faces control challenges due to system complexity and wind uncertainty. Accurate wind estimation at operational altitude is essential, yet current methods like extrapolation and EKF present significant drawbacks. This work proposes two novel, generalizable estimation techniques requiring only minimal sensor data (kite position, tether force). One employs optimization on a simplified model, while the second solves a linear system derived from specific dynamic assumptions, suitable for least-squares or Kalman filtering. Both methods are tested using real flight data and compared with surface wind speed measurements.

Keywords: Optimal Control, Aerospace, Machine Learning in modeling, estimation, and control
Abstract: In-Space or On-orbit Servicing, Assembly, and Manufacturing (ISAM/OSAM) is a growing field in developing space robotics to repair or refuel existing satellites, assemble large structures, and manufacture components in space. A foundational capability in this domain is the development of a control approach for OSAM vehicles to safely egress from the servicing or assembly area. This work compares the performance of model-based optimal control techniques to a Reinforcement Learning (RL) approach for the egress problem. Simulation results are presented where control efficiency is used as a primary metric for comparison.

Keywords: Assistive and Rehabilitation Robotics, Biomechanical Systems, Human-Machine and Human-Robot Systems
Abstract: As wearable robotic assistive devices become increasingly prevalent, it is essential to not only improve their physical performance but also deepen our understanding of how they affect the human body. This talk presents two interconnected projects that address this dual challenge. The first introduces a physics-based evaluation framework that estimates biomechanical metrics—such as joint loads, muscle forces, and fatigue—using a combination of exoskeleton sensor date and musculoskeletal modeling, eliminating the need for external motion capture or EMG systems. This integrated evaluation approach enables real-time, on-board, continuous monitoring of user-device interaction, offering a scalable solution as wearable devices grow in complexity and use.

The second project features the design and development of a novel shoulder exoskeleton that provides bilateral, decoupled assistance using a single actuator. This is achieved through a unique cam-driven, floating motor system that dynamically adjusts cable tensions to support overhead tasks at varying shoulder angles. This mechanical innovation is paired with a biomechanical study showing reduced activation in key shoulder muscles, such as the deltoids and supraspinatus.

Together, these projects demonstrate a shift in how assistive devices are designed and evaluated. Beyond providing physical assistance, future devices must incorporate intelligent sensing and modeling to understand how they are affecting the human user. By embedding evaluation tools directly into exoskeletons and validating them through novel hardware applications, this work bridges the gap between augmentation and insight—laying the foundation for next-generation wearable robotics.

Keywords: Assistive and Rehabilitation Robotics, Robotics, Control Applications
Abstract: Providing endless possibilities for control and assessment, robotic exoskeletons allow for many undiscovered human-robot interactions to further benefit movement therapies for neuromotor impairment. The nature of direct feedback and repeatability allows for intelligent control to modulate rehabilitation through improved assessment, increased repetitions, engaging training sessions, and individualized assistance. Affecting millions, neurological impairments, such as spinal cord injury, traumatic brain injuries, or cerebral vascular accidents, degrade quality of life and may lead to chronic symptoms. Impaired individuals may not have sufficient strength or motor control in regards to reaching activities and require movement rehabilitation to regain semblance of functionality. A critical role in reaching tasks and nervous system reorganization is increased repetitions with sufficient intensity of coordinated limb movements. Muscular functional electrical stimulation helps facilitate increased coordinated movements assisting during low volitional muscular control. However, these methods lead to muscular fatigue and limit the duration of rehabilitation training sessions. Combining muscular electrical stimulation with the benefits of robotic exoskeletons allow for mitigation of muscular fatigue and other unmodeled dynamics to provide increased rehabilitative exercises. Using deep neural networks to join these rehabilitative modalities allow for safe control with Lyapunov-based stability guarantees to further allow for increased coordinated movement tasks. Deep neural networks provide learned patient treatment, which targets an individualized point of care to better adapt coordinated movement therapy to each individual. Functional electrical stimulation, conjoined with an upper-extremity exoskeleton, allows for safe human robot interventions for coordinated movement therapies from tailored control methods via Lyapunov-based deep neural network control.

Keywords: Automotive Systems, Estimation, Machine Learning in modeling, estimation, and control
Abstract: Incorporating teleoperation in autonomous vehicles (AVs) will allow for effective human intervention during autonomy failures, thus ensuring safer deployment of AV technology. An important requirements for AV teleoperation is that the environment around the remote ego-vehicle needs to be recreated at the teleoperator station using camera images received over a wireless network. However, transmitting large camera and Lidar images over a wireless transmission suffers from significant problems of latency and bandwidth. In this talk a Predictive Display (PD) system which compensates for such latency and enhances AV teleoperation will be described. The PD system is based on estimating the position and orientation of the ego vehicle and of other nearby vehicles using nonlinear observers. The observer utilizes IMU, GNSS and radar sensors to perform accurate state estimation and vehicle tracking. These estimates are then used to transform delayed camera videos to create updated videos for display to the teleoperator. Image processing is done using both deep-learning methods and traditional computer vision techniques. The first part of the talk will present a human subjects study to compare teleoperation performance with and without PD. The results clearly demonstrate that even a 0.5 seconds delay in camera images can make it impossible to control the vehicle but the use of the developed PD system can enable safe remote vehicle control with almost as accurate a performance as the delay-free case. The second part of the talk elucidates the application of PD systems on real-world video data using many different algorithms. First a deep-learning based algorithm will be described which uses 3D reconstruction and image-inpainting to generate predictive video. Then a vector field based PD system will be described which uses estimated vector fields to synthesize new images from the delayed ones. The algorithms use new motion models for vehicle tracking, and novel nonlinear observers along with sensor fusion of multi-camera and Lidar data. The experimental data proves the superiority of the developed PD methods as compared to other state-of-the-art video prediction methods.

Keywords: Automotive Systems, Transportation Systems
Abstract: With the prevalence of autonomous vehicles in modern day traffic systems, it is possible that hackers may infiltrate a subset of these vehicles and change their driving parameters. These hacked vehicles, referred to as malicious vehicles, can be arbitrarily interspersed with the other (normal) vehicles and perform a series of coordinated, subtle velocity changes, with the objective of introducing undesirable waves in the traffic system, which can impact the overall vehicle flow and even fragment the road connectivity. Modelling individual vehicles in a traffic system can reveal important information about individual driving behavior and the impact of malicious vehicles, however the computational complexity increases considerably with the number of vehicles considered in the system. On the other hand, macroscopic models, which draw inspiration from Eulerian fluid models, describe the behavior of the traffic by describing lumped or aggregate quantities like density and average velocity, calculated at spatio-temporal intervals. Such models are highly capable of describing collective transport phenomena such as the evolution of congested regions or the velocity of propagation of traffic waves in a computationally efficient manner. In this talk, normal and malicious vehicles are modeled using a two-species macroscopic Partial Differential Equation (PDE) traffic model. The two-species model is analyzed using a combination of analytical and machine learning methods to detect the presence of malicious vehicles in the traffic, as well as quantify their number, distribution and impact on the traffic system. Non-linear PDE simulations highlight the efficacy of these analyses.

Keywords: Cognition modeling, Assistive and Rehabilitation Robotics, Human-Machine and Human-Robot Systems
Abstract: Intelligent Tutoring Systems (ITS) emulate human tutors by closing-the-loop between human learners and tutoring agents. However, in comparison to ITS developed for traditional disciplines (e.g., mathematics or language), developing ITSs for psychomotor skills has its own set of challenges. For example, assessing “correctness” of an answer to a math problem does not directly translate to evaluating correct psychomotor task performance. Key challenges for psychomotor ITS include creating a task knowledge space, personalizing agents to learner characteristics, and maintaining learner motivation. To address these design challenges, we propose a cognitively aware ITS for psychomotor learning with specific consideration of a task where users learn to safely land a quadrotor in a 2D simulator. Cognitive factors such as self-confidence and workload influence learners' self-efficacy and learning outcomes, yet their operationalization in psychomotor ITSs remains limited. In response to this, we train a within flight automation assistance algorithm based on an optimal control policy designed not only to calibrate self-confidence to performance but also calibrate workload to appropriate levels throughout the sequence of landing attempts. We design and leverage a learning stage classifier to quantitatively characterize novice-to-expert performance, bridging qualitative and quantitative learning stage representation. Our policy is trained using reinforcement learning methods with self-confidence, workload, and learning stage Markov Decision Process models. By combining learning stage classification, task performance metrics, and automation assistance, our system generates tailored formative feedback—positive, neutral, or negative—enhancing personalization. This approach addresses critical challenges of psychomotor ITS design, offering a framework for effective ITSs.

Keywords: Cyber physical systems
Abstract: Increased connectivity and automation in modem infrastructures have amplified the vulnerability against cyberattacks that can impair system operation. Rapid detection as well as isolation of these cyberattacks is crucial to ensure reliable and safe operations by dispatching appropriate targeted remedial measures. However, the problem of isolating the attack source based on system model and measurement is inherently ill-posed and often relies on the availability of redundant sensors. In this talk, I will present a data-driven strategy for detection-isolation of cyberattacks on actuation and sensor via the Koopman operator. Specifically, I will cover how the algorithm leverages the changes in Koopman modes and eigenfunctions to detect as well as isolate actuation and sensor cyberattacks, without prior knowledge of the system model or requirement for redundant sensing. The algorithm adopts an online small-data learning strategy that ensures resource-efficient implementation and broad generalizability while addressing the limitations of both traditional model-based and data-driven approaches. The algorithm is applied for attack detection-isolation in compromised electric vehicle charging, where the charging actuation commands and the sensor measurements from the vehicle battery can be corrupted. Finally, I will present case studies with high-fidelity battery simulations using 'PyBaMM' and 'liionpack' under rate-limited measurements and realistic uncertainties to demonstrate the efficacy of the proposed algorithms. I will thereby highlight its potential impact on the safety of real-world energy infrastructure.

Keywords: Human-Machine and Human-Robot Systems, Manufacturing Systems, Modelling, Identification and Signal Processing
Abstract: Please see the document.

Keywords: Machine Learning in modeling, estimation, and control, Robotics, Human-Machine and Human-Robot Systems
Abstract: Ensuring safety in learning-enabled control systems is increasingly vital as neural networks become integral to robotics and other complex dynamical systems. This talk presents recent advances in the formal verification and synthesis of neural control barrier functions (neural CBFs) that guarantee safety in nonlinear systems represented by neural networks. I begin by introducing a verified neural dynamic modeling framework that employs Bernstein polynomial-based over-approximations, which enables real-time safe control and achieves significant speedup over the complete verifier while maintaining safety guarantees. Empowered by our recent neural network verification toolbox, I will then introduce symbolic derivative bound propagation techniques to verify neural CBFs with improved tightness and scalability of formal verification. These methods form the mathematical backbone of two frontier applications. First, in PDE boundary control, I show how neural boundary CBFs ensure constraint satisfaction for systems governed by unknown PDEs via neural operator modeling. Second, I extend these safety concepts to human-AI conversations, modeling dialogue context state transitions in large language models (LLMs) as controllable dynamical systems. Using neural CBFs, we conduct online safety steering that provably prevents LLMs from multi-turn jailbreaking attacks. By unifying methods across physical and symbolic domains, this line of work opens the door to verifiable dynamical safety in a wide range of autonomous systems— from robotic systems to conversational AI agents. The talk emphasizes verifiable forward invariance induced by neural barrier function and the integration of neural dynamics with control theory, demonstrating a scalable path toward trustworthy and safe AI-driven control.

Keywords: Manufacturing Systems, Adaptive and Learning Systems, Control Design
Abstract: Please see my submitted document.

Keywords: Manufacturing Systems, Path Planning and Motion Control, Robotics
Abstract: Please see my attached document.

Keywords: Mechatronic Systems
Abstract: Cooperative control for connected and automated vehicles (CAVs) is a critical component of next-generation intelligent transportation systems, aimed at enhancing traffic safety and management. However, traditional methods like one-dimensional (1D) vehicle platooning are fundamentally limited, as they ignore lateral vehicle interactions and cannot utilize the full capacity of multi-lane roads. This work introduces a novel cooperative control framework, termed "vehicle flocking," designed to overcome these limitations by coordinating the planar motion of multi-agent CAVs as a cohesive flock. A key challenge is applying flocking control within complex, varying, and structured road environments where vehicle behavior is governed by human-defined traffic regulations. To address the challenge, novel vehicle flocking rules are defined that integrate vehicle dynamics, traffic regulations, and permanent road boundaries directly within the three foundational flocking principles: separation, alignment, and cohesion. This integration is achieved through several technical innovations, including a new elliptical lattice-based spacing policy, an artificial potential function to enforce road boundaries, and an artificial flow guidance method to navigate the multi-agent CAVs. The framework's power and versatility are demonstrated in three critical applications: (1) a virtual vehicle-based model for seamless ramp merging, (2) a field-of-view neighbor selection rule for robustness against time delays; and (3) a novel dual-loop MPC structure for systematic parameter tuning. Simulation results confirm this control strategy enables reliable, coordinated motion of multi-agent CAVs, allowing them to maintain stable formations while enhancing overall traffic efficiency.

Keywords: Mechatronic Systems, Robotics, Manufacturing Systems
Abstract: Multi-process, multi-material additive manufacturing has the potential to revolutionize the design and fabrication of complex electronic and electromagnetic systems. In this work, advances in multi-process, multi-material desktop 3D printing for electromechanical actuators and frequency selective surfaces are presented. First, novel 3D-printed axial- and radial-flux motors are introduced. These motors are fabricated with three- or four-axis 3D printers using silver direct ink write (DIW) and plastic fused filament fabrication (FFF). Multi-process printing techniques enable the printing of high efficiency stators and electromechanical actuators and grippers during a single print session within a single additive manufacturing system. This novel technology transforms on-demand manufacturing, enabling seamless in-field component replacement and system upgrades for components and actuators. Next, we explore the fabrication of a doubly-curved, multi-layer frequency selective surface (FSS) sub-reflector for satellite communication with a multi-process desktop 3D printer implementing non-planar printing techniques. With additive manufacturing, it becomes possible to manufacture a complex, non-planar FSS with doubly-curved and parabolic shapes. Here, a triband FSS sub-reflector is designed and printed to expand satellite communication capabilities by separating S and C (2 to 5 GHz) wireless frequency bands from highspeed Ku (17.7 to 20.2 GHz) and Ka (27.5 to 30.0 GHz) frequencies without the need for separate antenna systems. This reduces satellite weight, cost, and complexity. Finally, a pick and place capability integrated within multi-process additive manufacturing is explored, including a demonstration of an axial three-phase electric motor with integrated Hall effect sensor circuitry. Overall, the multi-process additive manufacturing techniques described in the work enable rapid, on-demand prototyping and the manufacturing of mission-critical parts for applications that are beyond the reach of traditional supply chains. These innovations unlock new possibilities for space exploration, resilient infrastructure, and next-generation wireless technologies.

Keywords: Modeling and Control of Biotechnological Systems, Healthcare systems, Modelling, Identification and Signal Processing
Abstract: Acute mental stress arises in many everyday life circumstances in response to perceived threats (i.e., stressors) and negatively impacts the psycho-physiological states of individuals. It has been shown that acute mental stress induces sympathetic arousal, whose frequent occurrence may deteriorate the quality of life. This talk covers our recent studies on transcutaneous median nerve stimulation (tMNS) as an emerging non-invasive modality used for its therapeutic effect on mitigating stress-induced sympathetic arousal. We share our experimental results and developments toward a tMNS-enabled personalized health intervention from the three standpoints of i) sensing, detection, and monitoring, ii) modeling cardiovascular responses to acute stress and tMNS, and iii) closed-loop just-in-time interventions. First, we present a novel acute mental stress detection algorithm based on statistical inference and a novel synthetic multi-modal variable (SMV) that enables real-time and personalized monitoring of cardiovascular responses to acute mental stress and tMNS. The SMV integrates six plausibly explainable physio-markers extracted from three wearable sensing modalities, namely photoplethysmogram (PPG), electrocardiogram (ECG), and seismocardiogram (SCG). The experimental data from healthy individuals suggest that the SMV exhibits superior consistency, sensitivity, and robustness compared to individual physio-markers. We also present an inference-enabled algorithm that leverages the SMV to track and monitor stress-induced cardiovascular arousal. Second, we present a virtual experiment generator (VEG) developed through system identification and variational inference approaches to replicate population-level and subject-specific cardiovascular responses to acute mental stress. VEG enables in silico testing and development. Third, we leverage the VEG and design and evaluate robust control algorithms for just-in-time closed-loop controlled tMNS to effectively mitigate acute stress-induced arousal despite large inter- and intra-individual variability. Overall, this talk presents the system-theoretic challenges and opportunities of developing closed-loop tMNS interventions for mitigating stress-induced sympathetic arousal.

Keywords: Modeling and Validation, Manufacturing Systems, Control Design
Abstract: Please see the document.

Keywords: Modelling, Identification and Signal Processing, Optimal Control, Control Design
Abstract: Proton Exchange Membrane (PEM) fuel cells are a leading clean energy technology for transportation and stationary applications due to their high efficiency and low emissions. Their performance, however, is highly dependent on maintaining optimal membrane hydration, which is governed by internal humidity conditions. This study develops control-oriented models and strategies to manage humidity within PEM fuel cells. A lumped-parameter physics-based model was developed in MATLAB/Simulink to predict internal humidity dynamics. While grounded in physical laws, such models can struggle with accuracy due to assumptions and the complexity of PEM fuel cells. To address this, data driven modeling techniques were explored. In particular, the Koopman operator, a linear but infinite-dimensional operator capable of capturing nonlinear dynamics, was used. Finite dimensional approximations were constructed using time-delay embeddings and radial basis functions. A NARX neural network was also investigated and compared to the Koopman model. Among the models studied, the Koopman model with time-delay embeddings outperformed both the physics-based and NARX models, showing strong potential for capturing nonlinear water management behavior in PEM fuel cells. This Koopman model was then employed in a model predictive control framework (KMPC) as the predictive model, while the physics-based model served as the plant to optimize anode relative humidity and performance in an open-cathode PEM fuel cell stack. The performance of KMPC was compared to a baseline proportional-integral (PI) controller. While both achieved similar reference tracking, KMPC adjusted control effort based on operating conditions, improving efficiency. Additionally, KMPC’s linearity enables the use of efficient linear MPC solvers, making it well-suited for real-time control.

Keywords: Motion and Vibration Control, Manufacturing Systems
Abstract: Electrohydrodynamic (EHD) printing has become a promising and cost-effective technique for producing high-resolution and large-scale features. One widely recognized obstacle in EHD printing is nozzle clogging due to solvent evaporation or ink polymerization. Moreover, printing highly viscous materials often requires pressure or other external force to assist the ink flow during the printing, which increases the complexity of process control and the required energy. In this work, we developed a novel ultrasonic vibration-assisted EHD printhead and associated process to effectively eliminate the nozzle clogging for the printing of high-viscosity and high-evaporation-rate inks. A series of experimental tests were conducted to characterize the printhead design and process parameters (i.e., vibration frequency, vibration amplitude, and printing voltage). The results demonstrated that superimposing ultrasonic vibration on the EHD printing nozzle can effectively enhance current EHD printing capabilities, such as reducing required pressure, eliminating nozzle clogging, and providing stable and continuous printing for high viscosity and high solvent evaporation rate material. With the optimal parameters, a filament with a diameter of around 1μm can be continuously printed, and we successfully applied this developed ultrasonic-assisted EHD process to print high-resolution 2D patterns.

Keywords: Optimal Control, Control Applications
Abstract: Manufacturing industries face increasing pressure to improve sustainability while maintaining production efficiency and meeting customer demands. As renewable energy integration accelerates and electricity markets shift toward dynamic pricing structures, manufacturing systems must evolve beyond traditional scheduling paradigms to become grid responsive. However, at scale, these energy-aware manufacturing control schemes must be responsive to 1) time-varying electricity prices and renewable energy availability, 2) complex networked production flows and inventory constraints, and 3) customer order requirements and production targets. This talk will present recent results on hierarchical model predictive control methods that enable manufacturing systems to be both customer-responsive and energy-responsive. We will explore how networked manufacturing system models can capture complex material flows and energy consumption patterns across interconnected machines and buffers. The presentation will demonstrate how model predictive control frameworks can dynamically optimize production schedules in response to real-time pricing while maintaining production commitments. Finally, we will examine bi-level optimization approaches that jointly optimize product pricing and production scheduling, enabling manufacturers to shape demand while leveraging renewable energy availability. Through case studies involving battery manufacturing systems, we will show how these integrated approaches can achieve significant energy cost reductions while maintaining profitability and production targets. The talk will highlight the potential for manufacturing systems to become active participants in demand-side energy management, contributing to grid stability while enhancing their own operational efficiency and sustainability.

Keywords: Power and Energy Systems, Transportation Systems
Abstract: This presentation surveys the literature on barrier function-based safety control and identifies a critical research gap, namely, the need for safety control algorithms that degrade gracefully in the presence of system failures and anomalies. The presentation proposes a novel control design paradigm that embeds the notion of graceful degradation within the control barrier function theory. Intuitively, the idea is to switch from one safety control mode to another in response to changes in the system's environment and/or damage/hazard state. Mathematically, we achieve this by modifying the design of the barrier functions and constraints to create a multi-layered definition of safety. We illustrate this approach for two different application problems, namely, preventing thermal runaway propagation in lithium-ion batteries and ensuring graceful collision avoidance in adaptive cruise control applications.

Keywords: Robotics
Abstract: Autonomous systems are rapidly reshaping domains like transportation, manufacturing, and everyday human life. Yet, their widespread deployment hinges on a critical question: Can we ensure these systems consistently behave as intended—never in ways we cannot predict or control? This talk presents a cohesive research agenda that tackles this challenge through three interconnected thrusts: (i) expressive task specification, (ii) correct-by-construction planning and control, and (iii) formal robustness certification.

I begin by introducing hierarchical temporal logic specifications that express rich combinations of safety and liveness goals while remaining intuitive for engineers to write—and even translatable from natural language via large language models. These representations dramatically reduce specification time and scale to complex multi-robot coordination tasks.

Next, I highlight a suite of planning, control, and decision-making algorithms that rigorously fulfill such specifications. This includes the first abstraction-free motion planner with both probabilistic completeness and asymptotic optimality, scalable task allocation and motion planning for multi-robot collaboration, and methods that leverage previously solved problems through plan reuse.

Finally, I demonstrate how these guarantees extend into modern learning-based components. In collaboration with Boeing, I present the first formal certification of 6-DoF pose estimation pipelines built on keypoint detection, even under challenging visual conditions such as occlusion and more complex semantic perturbations.

Combined with the principles of specification, synthesis, and verification, these contributions lay the foundation for autonomous systems that are both capable and backed by formal guarantees of reliability.

Keywords: Robotics
Abstract: Robotics foundation models pre-trained on internet-scale data have started to revolutionize how robots understand the complex world, interpret human feedback, and plan actions. The promise of using these models within robotics is the ability to generalize robot behaviors and push robot deployment into increasingly unstructured or novel environments. However, despite the remarkable progress, it is precisely this integration of foundation models that introduces new safety and alignment challenges in robotics. Robots are safety-critical systems, wherein a foundation model’s single erroneous visual or language interpretation, misaligned behavior generation, or high inference latency can lead to catastrophic consequences. In this talk, I will introduce our recent efforts to generalize model-based control principles to tackle the challenges that have emerged throughout the lifecycle of robotics foundation models, ranging from pre-training to post-training and deployment. On the pre-training side, I will introduce our work on identifying informative “system-level” model failures for cost-effective and targeted model training. On the post-training side, I will introduce our efforts to bring the success of preference alignment, widely adopted in non-embodied foundation models (e.g., large language models), to embodied contexts such as robot manipulation and autonomous driving, enabling robots to align their behavior with human preferences with minimal human feedback. Finally, on the deployment side, I will share our work on test-time model monitoring and adaptation for safe deployment of robotics foundation models in the novel environments.

Keywords: Robotics, Mechatronic Systems, Soft Robotics
Abstract: This talk presents a structure-driven approach to robotic grasping and manipulation, in which key control and modeling challenges are simplified through mechanical design innovation. We introduce mechanical closure, extending classical closure definitions via adaptive structures such as multi-segment soft actuators and origami-based grippers. These enable delicate and robust grasps in complex scenarios—from capturing jellyfish underwater to handling raw egg yolks. To simplify contact uncertainty, we propose adaptive variable stiffness phalanges, allowing reliable interaction across diverse geometries and friction conditions. For in-hand manipulation, the DexCo Hand, a soft–rigid hybrid system with soft hydraulic actuation, reduces control burden through intrinsic compliance. Building on this foundation, we further incorporate selective learning-based frameworks that transfer human-inspired skills—such as the autonomous in-hand manipulation required to in-hand counting and sorting pills. These systems demonstrate how embedding intelligence into mechanical structure, with minimal sensing and model dependence, leads to robust, dexterous, and scalable manipulation. My approach brings these three elements—structure, control, and AI—into synergy, instead of opposition.

Keywords: Robotics, Path Planning and Motion Control, Optimal Control
Abstract: Set-based methods are widely used in control analysis and design. Example applications in robotics include verifying the safety of system trajectories and defining constraints to which a controller must adhere. In the context of reachability analysis, zonotopic sets (i.e., zonotopes and their generalizations) have become popular in recent years because they have closed-form expressions for many important set operations and favorable complexity growth when compared to more traditional set representations. A downside to these sets is that analysis generally relies on numerical optimization.

In this talk, the coupled problems of designing zonotopic sets to be efficiently analyzed and designing optimization algorithms to analyze these sets are considered. Further, reachability analysis using zonotopic sets is conceptualized as a general methodology for building structured optimization problems for dynamic systems. Sparsity-promoting reachability calculations are presented that have lower memory complexity when compared to typical methods. Optimization problems built using these calculations are efficiently solved using tailored approaches based on the alternating direction method of multipliers (ADMM). Optimization times are shown to be faster than state-of-the-art solvers and problem formulations. This reachability analysis and structured optimization workflow can be implemented online to facilitate real-time control, estimation, and safety verification. The proposed framework is applied to both convex (using constrained zonotopes) and non-convex (using hybrid zonotopes) problems, such as those involving hybrid systems and disjoint constraint sets. New methods are experimentally validated in application to path and motion planning for ground robots.

Keywords: Soft Robotics, Mechatronic Systems, Sensors and Actuators
Abstract: Soft robots are increasingly utilized for exploring unstructured environments and delicately handling objects, primarily due to their inherent compliance and safe interaction with humans. To enable soft robots with embodied perception, it is essential to acquire accurate proprioceptive feedback and external environmental perception (exteroception). However, unlike traditional rigid robots equipped with precise rigid sensors like encoders, soft robots face significant challenges in proprioception due to their infinite degrees of freedom. Additionally, it is crucial for soft robots to estimate external forces without exterior rigid sensors when interacting with the environment in applications such as minimally invasive surgery. This talk presents an integrated approach to simultaneous continuous shape reconstruction and external force estimation for soft robots using a novel distributed inductive curvature sensor. This sensor captures continuous actuator shapes in real-time through electromagnetic induction, enabling high accuracy and practical scalability. In addition, an enhanced analytical model based on the Euler–Bernoulli curved beam theory is developed to predict the shape under pneumatic actuation and external forces. External forces are estimated through a model-based optimization approach based on the measured shape. Furthermore, this talk will also introduce a learning-based modeling framework leveraging a Euler spiral-inspired shape representation for soft robots. Using this representation, we develop neural network-based forward and inverse models that accurately predict the actuator's shape and estimate external forces. In summary, this talk presents advances in embedded sensing and learning-based modeling that significantly enhance embodied perception—integrating proprioception and exteroception—and pave the way for more intelligent, adaptive, and autonomous soft robotic systems.

Keywords: Soft Robotics, Modeling and Validation, Robotics
Abstract: Soft continuum robots offer benefits compared to conventional rigid robots for applications in medical settings, grasping/handling objects, and navigating confined and unstructured environments due to their inherent compliance. A highly sought after goal is to replicate the dexterity and manipulability of biological structures, such as elephant trunks. However, the continuous form and underactuation of soft robots pose challenges in modeling and control and is an active research area. In this talk, I will describe a modeling framework based on a constrained Lagrangian approach in non-minimal coordinates that can accommodate holonomic and non-holonomic constraints and capture the dynamic characteristics of soft continuum robots. The simplicity and generalizability of this approach is unique and is currently not replicated in the soft robot modeling literature. To demonstrate the model's efficacy, we designed and validated two highly dynamic soft robots. The first is biologically-inspired and mimics the motion of a Pacific-lamprey fish. It is able to scale a burlap-lined vertical wall up to 10 cm/sec (0.47 body lengths per second, blps). The second is a wheeled robot that is able to locomote over a planar surface up to 22 cm/sec (1.38 blps), reverse, and perform circular motion. Both these robots are modeled with this framework and their configuration and velocities are characterized over a broad frequency range (0.5 to 5 Hz) demonstrating a close match with simulation. We are currently extending this open-loop validation for closed-loop, model-based control of a 2-DOF soft actuator. The soft actuator is designed to enable grasping and facilitate environmental interaction by increasing curvature along its backbone to the tip. Preliminary simulation results indicate robustness to parametric uncertainty and low tracking error for line and circle following tasks at relatively high frequency. The implementation of real-time robust control for this class of highly dynamic underactuated soft actuators will significantly improve soft robot functionality.

Keywords: Biomechanical Systems, Machine Learning in modeling, estimation, and control, Sensors and Actuators
Abstract: Essential tremor (ET) is a neurological disorder that causes involuntary rhythmic limb movements, negatively impacting quality of life. ET’s severity is quantified by The Essential Tremor Rating Assessment Scale (TETRAS). To extend TETRAS, a custom Long Short-Term Memory (LSTM) network using two connected LSTM layers with dropout in between, followed by a fully connected layer and a scaled sigmoid regression output with global average pooling was developed to classify ET with Inertial Measurement Unit (IMU) data collected from twelve participants previously diagnosed with ET. Training employed two data segmentation methods: one designed to simulate the constraints of TETRAS, and one using a traditional data split. Both were first evaluated using a K-fold cross-validation with a K of 10 to evaluate model performance on the small dataset, resulting in an average validation mean squared error (MSE) of 31.4530 ± 9.4181 and 31.7982 ± 23.5870, respectively. The segmentation methods were then used to train two models: one using a 70/20/10 training, validation, and test split; and one with three participants removed and used as unseen test data, which resulted in a final validation MSE of 54.1002, RMSE of 5.6471, and R2 of 0.633 for the random split and a 20.5013, 4.5278, and 0.226 for the traditional split. An RMSE of ∼5.6 represents less than 10% of the TETRAS range, indicating that the observed error falls within the established clinical variability.

Keywords: Healthcare systems
Abstract: This paper investigates the robustification of the population-informed-personalized Gaussian sum extended Kalman filter (PI-P-GSEKF) developed in our prior work and its application to continuous venous oxygen saturation (SvO2) estimation. The PI-P-GSEKF was developed to enable state estimation in systems with extremely large variability. It includes a bank of EKFs, whose operating points (i.e., nominal parameter vectors) are selected via generative sampling followed by Markov Chain Monte Carlo (MCMC) sampling with one-time partial state measurement. The state is estimated as the weighted sum of state estimates from all the EKFs, with the weight for each EKF calculated based on the likelihood of its prediction at every measurement instant. Despite its adequate performance in general, its state estimate can suffer from high-sensitivity operating points whose inaccuracy with respect to the ground truth operating point results in large EKF errors and adversely impact the PI-P-GSEKF. We explored two ideas to robustify the PI-P-GSEKF against this challenge: (i) penalizing high-sensitivity operating points in MCMC sampling (called robust MCMC sampling) and (ii) calculating the weights for the EKFs based on the likelihood of their predictions in a measurement horizon (called robust Gaussian summing). We examined the efficacy of these ideas in the context of continuous SvO2 estimation from arterial oxygen saturation (SpO2) measurement, which is important in critical care and cardiopulmonary medicine but is highly invasive and challenging. The results suggested that both ideas could reduce SvO2 estimation error compared with the standard PI-P-GSEKF ((i): 4%; (ii): 13%; (i)+(ii): 16%, all on the average). However, how to set the length of the sampling interval for weight calculation remains an open challenge.

Keywords: Healthcare systems
Abstract: Cardiovascular arousal to acute mental stressors poses significant risks to health and contributes to cardiovascular and neurological disorders. Transcutaneous median nerve stimulation (TMNS) is an emerging non-invasive intervention with the potential to enable just-in-time mitigation of cardiovascular arousal to acute mental stressors. However, no prior work has addressed how to best control TMNS for this purpose. A critical limitation is the lack of basic knowledge of the dynamics between acute mental stressors and TMNS versus cardiovascular arousal. As an initial step toward investigating closed-loop controlled just-in-time TMNS for mitigating cardiovascular arousal to acute mental stressors, we developed and evaluated a data-driven virtual experiment generator (VEG) which can replicate the dynamics between acute mental stressors and TMNS versus cardiovascular arousal. In terms of a novel synthetic multi-modal variable (SMV) intended to serve as feedback for just-in-time TMNS, we derived a 4th-order linear decoupled multi-input-single-output representation to replicate the dynamics between acute mental stressors and TMNS as inputs versus SMV as output using data collected from 23 experiments. The representation could (i) replicate SMV responses in all experiments when parameterized with the experiment-specific parametric probability density functions (PDFs) and (ii) generate plausible virtual experiments when parameterized with the experiment-aggregated parametric PDFs. In sum, the VEG has the potential as a virtual platform to develop and evaluate closed-loop controlled just-in-time TMNS for mitigating acute mental stressor-induced cardiovascular arousal with SMV as feedback.

Keywords: Healthcare systems, Biomechanical Systems, Mechatronic Systems
Abstract: This paper presents an analysis of the precision of fingertip location for a 7 degrees of freedom (DOF) pneumatic haptic glove developed at SMU for 3D elastographic imaging virtual palpation. The 2D workspace for each finger and the 3D workspace for the glove are calculated from the actuator positions using closed-form expressions based on design geometry. The fingertip locations are measured using a 5-camera OptiTrack motion capture, and the results are compared with the fingertip positions obtained using the linear sensors integrated in the design of each finger joint. This work is an essential step in the development of a virtual palpation system of 3D elastographic medical images.

Keywords: Healthcare systems, Estimation, Cyber physical systems
Abstract: This paper develops methods to estimate breath-by-breath tidal volume using wearable inertial measurement unit sensors located on the chest and abdomen of the subject. Measurement of tidal volume is typically done in a clinic using a nose clip and having the subject breathe through a spirometer held in the mouth. The new methods developed in this paper offer significant advantages for patient groups unable to perform or tolerate such a technique, and also enable continuous home-based monitoring of respiratory variables. First, a method is developed that estimates chest and abdominal displacements during respiration, followed by constructing a linear regression model to relate tidal volume to these displacements. In this case, a nonlinear observer is utilized to estimate respiratory displacements by the accelerometer and gyroscope signals of the IMUs. Next, an alternate method based on the use of a deep learning CNN-LSTM model to directly estimate tidal volume from raw IMU signals is developed. In this case, the estimation of intermediate displacements is not necessary. However, it requires significantly higher computational power. Both methods achieve an overall accuracy of the order of 0.1L in tidal volume estimation for each breath over tidal volumes range of ~400 to 1500 mL.

Keywords: Healthcare systems, Machine Learning in modeling, estimation, and control
Abstract: Acute coronary syndromes (ACS) are a leading cause of morbidity and mortality worldwide, emphasizing the urgent need for reliable prediction of major adverse cardiovascular events (MACE) to support timely clinical decision-making. This study develops and validates machine learning (ML) models for MACE prediction using retrospective data from 2,721 ACS patients. A total of 109 predictor variables were extracted, encompassing demographics, comorbidities, and clinical parameters. A natural language processing (NLP) scheme was developed to transform the unstructured clinical text into quantitative scores. Five MLs, including Random Forest (RF), Support Vector Machine (SVM), k-Nearest Neighbors (kNN), Neural Network (NN), and Logistic Regression (LR), were developed to predict MACE across 30-day, 1-year, and 3-year horizons. RF consistently outperformed other models with Area Under the Curves (AUCs) ranging from 0.82 to 0.87, and achieved the highest aggregated performance rank across six evaluation metrics. This performance surpasses that of prior studies, which reported AUCs between 0.71 and 0.81. Cohort analysis revealed a progressive increase in MACE prevalence and consistent clinical differences between MACE and non-MACE patients, including older age, comorbidities, and abnormal vital and laboratory values. The RF’s strong performance supports its clinical utility in both acute triage and long-term cardiovascular risk monitoring. The proposed framework holds promise for enhancing patient outcomes, optimizing healthcare resource allocation, and reducing costs through informed and timely interventions.

Keywords: Marine Systems, Nonlinear Control Systems
Abstract: This study focuses on the tracking control problem of an unactuated surface vessel guided by six uni-directional tugboats. The control problem is further complicated as the velocity of the center of mass of the unactuated ship is considered to be unavailable. Specifically, a nonlinear model-free velocity observer formulation is designed to eliminate the reliance on direct velocity measurements in the control formulation. A model-based nonlinear control strategy is then developed based on the estimated velocity to achieve accurate trajectory tracking. The closed-loop stability of the system is rigorously analyzed using Lyapunov-based techniques. The proposed method guarantees the asymptotic convergence of the tracking error to the origin. The effectiveness of the proposed control approach is validated through comprehensive numerical simulation studies.

Keywords: Modeling and Validation, Marine Systems
Abstract: This study presents key enhancements in hydrodynamic modeling using the strip-based Morison’s equation approach to enable rapid simulations of floating offshore wind turbines (FOWT). The modeling framework employs the relative form of the Morison equation, incorporating nonlinear irregular wave kinematics, vertical wave stretching, and diffraction corrections based on MacCamy-Fuchs (MCF) theory for large-scale, non-slender structures. Wave kinematics are iteratively applied at dynamically displaced structural nodes to accurately capture fluid-structure interaction. Additionally, a discretization scheme is introduced to improve hydrodynamic load distribution across large horizontal structures of floaters. These enhancements are validated against experimental data from the Floating Offshore Wind and Controls Advanced Laboratory (FOCAL), which conducted a 1:70 scale test of the IEA-Wind 15MW reference turbine on the VolturnUS-S platform. Results demonstrate that the incorporation of nonlinear wave kinematics significantly improves low-frequency response accuracy. Furthermore, the vertical wave stretching and MCF corrections lead to surge response predictions that closely align with experimental measurements, while the improved load discretization significantly enhances heave and pitch response fidelity in wave-dominant frequency ranges.

Keywords: Modeling and Validation, Marine Systems, Large Scale Complex Systems
Abstract: This paper presents the modeling and verification of multibody structural dynamics for offshore wind turbines. The flexible tower and support structure of a monopile-based offshore wind turbine are modeled using an acausal, lumped-parameter, multibody approach that incorporates structural flexibility, soil-structure interaction, and hydrodynamic models. Simulation results are benchmarked against alternative modeling approaches, demonstrating the model's ability to accurately capture both static and dynamic behaviors under various wind and wave conditions while maintaining computational efficiency. This work provides a valuable tool for analyzing key structural characteristics of wind turbines, including eigenfrequencies, mode shapes, damping, and internal forces.

Keywords: Modeling and Validation, Power and Energy Systems, Marine Systems
Abstract: This paper presents a novel approach for implementing frequency-dependent hydrodynamic coefficients in Morison’s equation, which is widely used in hydrodynamics modeling. Accurate hydrodynamic predictions using Morison’s equation necessitate the incorporation of frequency- dependent drag coefficients due to their variation with wave frequency. To address this, the proposed method segments the frequency domain into different regions, such as low-frequency (resonance) and high-frequency (wave) regions. Instead of using a constant drag coefficient across the entire spectrum, different drag coefficients are assigned to these regions. To implement this, a fifth-order low-pass Butterworth velocity filter is applied for the resonance zone, while a first-order high-pass Butterworth velocity filter is applied for the wave-dominated zone. The approach is validated using the INO WINDMOOR 12MW semisubmersible offshore wind turbine, comparing the simulation results against the experimental data. By incorporating frequency-dependent drag coefficients, the model shows improved agreement with experimental surge motion data across both frequency regions, demonstrating the effectiveness of the proposed method.

Keywords: Nonlinear Control Systems, Power and Energy Systems, Control Applications
Abstract: Floating offshore wind turbines (FOWTs) offer advantages over their land-based counterparts through access to better wind conditions and fewer socio-economic constraints. The dynamics of floating platforms present engineering challenges and opportunities. Wind and wave disturbances induce motion of the FOWT that can reduce power quality and increase structural loads. In contrast, specific movements of the floating platform may be desirable, especially in wind farm wake control applications. This research develops analytical platform motion control laws using a reduced-order FOWT model with six degrees of freedom for platform motion and rotor spin. In particular, Lyapunov’s direct method is used to develop an unconstrained control law for platform translation and rotation control. A model based on the NREL-5MW reference turbine mounted on the OC3-Hywind spar-buoy reference floating platform is utilized to test the control under ideal actuator assumptions. The performance is validated in higher-fidelity simulations for steady, below-rated wind conditions with irregular waves. Possible use cases and applications using novel platform actuators are discussed for future work.

Keywords: Estimation, Underwater Vehicles, Multi-agent and Networked Systems
Abstract: This paper presents a relative localization technique for underwater swarm robotics using received signal strength (RSS) from 433 MHz RF transceivers with directional antennas, fused with inertial measurements from an onboard Inertial Measurement Unit (IMU) via an Extended Kalman Filter (EKF). A leader-follower scenario is simulated in a water tank environment, where the follower fish estimate their position relative to a leader using only RSS and IMU data. The proposed method achieves a reliable estimate of relative distance and orientation, enabling effective formation control. The simulation results demonstrate the feasibility of this approach for distributed underwater swarm coordination.

Keywords: Estimation, Optimal Control, Electromechanical systems
Abstract: Large-deformation microactuators, organized in origami-inspired panel/hinge architectures, can be used to construct a variety of mechanisms for small-scale manipulation. In such manipulation applications, it is often desirable to measure the interaction forces that arise between the actuation mechanism and the external object. At the micro-scale, force estimation can be challenging due to device variability, constraints on space for sensor integration, and limited sensing resolution. This paper presents a simulation study applying a model-predictive control architecture with two different penalty mechanisms to disturbance force estimation in a panel-hinge micro-structure. The proposed architecture is found to provide improved performance over Kalman filter-based approaches in the context of limited sensor resolution and sampling rate. The different penalty mechanisms offer different advantages with respect to either model parameters mismatch or output noise. Performance trends with model error are assessed, with discussion of practical considerations for implementation with prototype micro-origami structures.

Keywords: Estimation, Stochastic Systems
Abstract: We address the challenge of temporal change detection and estimation in nonlinear dynamical systems. Traditionally, model-based fault diagnosis methods rely on the parallel execution of multiple filters, each corresponding to different system models (nominal and faulty). Although effective, this approach becomes computationally expensive, especially when using numerical techniques such as the particle filter instead of closed-form, Gaussian filters. In this work, we propose a novel fault diagnosis framework that integrates a Bernoulli state variable, in the form of a jump Hidden Markov Model, governed by a task-specific probabilistic mode switching mechanism, to represent the presence or absence of a fault (or change) within a dynamic system. By embedding this binary state within a Bayesian filtering framework, we develop a hybrid state filter that significantly reduces the computational complexity traditionally associated with particle filtering approaches for fault diagnosis. The proposed filter enables real-time evaluation of fault hypotheses based on the statistical moments of a single posterior probability density function, thereby eliminating the requirement for maintaining multiple parallel filters. The proposed method is validated through a numerical benchmark example demonstrating its efficiency and effectiveness.

Keywords: Estimation, Stochastic Systems, Modelling, Identification and Signal Processing
Abstract: In this paper, we introduce a novel Fault Diagnosis (FD) strategy for autonomous systems. By leveraging an underlying dynamic model of a system, our approach combines a binary state change detector with the Particle Filter (PF) technique to realize a hybrid estimator that simultaneously estimates both binary and continuous states. This hybrid filter thus achieves concurrent estimation of fault modes (represented as fault probabilities over time) alongside continuous valued states, offering a simpler and more effective solution than traditional Multiple Model (MM) strategies for FD. Our method excels in real-time online monitoring while enabling the diagnosis of multiple simultaneous faults without increasing computational complexity. We validate our algorithm using a benchmark stochastic vehicle platoon kinematic model. The results demonstrate our algorithm's efficacy and suitability for FD of safety-critical systems with single or multiple concurrent faults.

Keywords: Estimation, Stochastic Systems, Nonlinear Control Systems
Abstract: This work introduces a new reduced order extended Kalman filter (REKF) for discrete nonlinear dynamic systems which is derived by local minimization of the filter estimation error covariance. The Kalman gain and Riccati equation are obtained by completing the square. Both the filter and the 1-step ahead predictor form are derived. For demonstration purposes the REKF is applied to a polynomial nonlinear system, and its estimation error is compared to that of the full-order EKF.

Keywords: Estimation, Unmanned Ground and Aerial Vehicles, Intelligent Autonomous Vehicles
Abstract: This paper deals with the topic of pose and sensor bias estimation based on GNSS and IMU measurements which is a widely discussed topic but usually by omitting the rigorous check of state observability leading to the suspicion of invalid assumptions. The current work underlines this suspicion by pointing out through observability calculations that to estimate position, velocity, attitude and acceleration and angular rate biases of a system, attitude related measurements should be included besides GNSS position and velocity data. Such measurement can be the magnetic vector. Theoretical results are underlined by tests considering real flight GNSS and IMU data of a DJI M600 Pro multicopter and comparing the results to the onboard DJI pose estimator.

Keywords: Adaptive and Learning Systems, Modelling, Identification and Signal Processing

Keywords: Automotive Systems, Power and Energy Systems, Modeling and Validation

Keywords: Human-Machine and Human-Robot Systems, Robotics, Mechatronic Systems

Keywords: Manufacturing Systems, Optimal Control

Keywords: Manufacturing Systems, Stochastic Systems, Path Planning and Motion Control

Keywords: Soft Robotics, Machine Learning in modeling, estimation, and control, Estimation

Keywords: Machine Learning in modeling, estimation, and control, Estimation, Transportation Systems
Abstract: The limited service life of onboard batteries for EVs is a challenge, underscoring the need for real-time battery usage prediction. This paper proposes an adaptive Algebraic Derivative Estimation (ADE) approach for forecasting the energy consumption of battery electric buses. By dynamically adjusting the sliding window length, the adaptive ADE retains the fixed-length ADE’s key advantage—namely, operating online without reliance on extensive historical datasets—while substantially bolstering forecast accuracy by actively trading estimation bias off estimation variance. Comparative experiments against both the conventional ADE with a fixed length and a representative machine learning algorithm, XGBoost, were conducted, with performance evaluated via root mean square error, mean absolute error, and the coefficient of determination. The results demonstrate that the proposed approach significantly outperforms baseline methods.

Keywords: Machine Learning in modeling, estimation, and control, Transportation Systems, Modeling and Validation
Abstract: Car-following models are fundamental components in traffic flow theory, essential for understanding and simulating vehicular interactions on roadways. They play a crucial role in traffic management, infrastructure planning, and the development of advanced driver assistance systems and autonomous vehicles. In this study, we propose a novel car-following model based on CatBoost and SHapley Additive exPlanations (SHAP) to achieve both high predictive accuracy and interpretability. We compare our model with the widely used Intelligent Driver Model (IDM) to evaluate its performance comprehensively. The results demonstrate that our CatBoost-SHAP model significantly outperforms the IDM, achieving lower Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) values, along with a higher R-squared (R²) score. Specifically, our model captures complex driving behaviors more effectively, leading to enhanced predictive capabilities. Furthermore, SHAP analysis reveals that Relative Velocity is the most influential factor affecting acceleration decisions. The integration of CatBoost with SHAP addresses the limitations of traditional models by offering a data-driven approach that is both precise and explainable, making it suitable for practical applications in traffic modeling and autonomous vehicle development.

Keywords: Intelligent Autonomous Vehicles, Machine Learning in modeling, estimation, and control, Transportation Systems
Abstract: Accurate motion prediction is essential for safe and effective control in Connected and Automated Vehicles (CAVs). While existing methods are primarily designed for short-horizon prediction to support safety-critical tasks, emerging energy efficiency and mobility applications require reliable long-horizon predictions across extended road segments. To address this gap, we propose a novel traffic flow aware motion prediction framework that leverages the spatiotemporal connection of sequential road segments. Our framework incorporates contextual information from neighboring scenes to enable more consistent and informed long-horizon predictions. Simulation results on the I-24 MOTION dataset validate the effectiveness of the proposed framework, showing substantial performance improvements when adjacent scene information is included, particularly at longer prediction horizons. Implementation code of the proposed framework is available at: https://github.com/shao-lab-uga/TFA.

Keywords: Estimation, Modelling, Identification and Signal Processing, Robotics
Abstract: Accurate modeling is crucial in many engineering and scientific applications, yet obtaining a reliable process model for complex systems is often challenging. To address this challenge, we propose a novel framework, reservoir computing with unscented Kalman filtering (RCUKF), which integrates data-driven modeling via reservoir computing (RC) with Bayesian estimation through the unscented Kalman filter (UKF). The RC component learns the nonlinear system dynamics directly from data, serving as a surrogate process model in the UKF’s prediction step to generate state estimates in high-dimensional or chaotic regimes where nominal mathematical models may fail. Meanwhile, the UKF’s measurement update integrates real-time sensor data to correct potential drift in the data-driven model. We demonstrate RCUKF’s effectiveness on well-known benchmark problems and a real-time vehicle trajectory estimation task in a high-fidelity simulation environment.

Keywords: Nonlinear Control Systems, Control Design
Abstract: In this paper, we present an analysis of the Unscented Transform Controller (UTC), a technique to control nonlinear systems motivated as a dual to the Unscented Kalman Filter (UKF). We consider linear, discrete-time systems augmented by a bounded nonlinear function of the state. For such systems, we review 1-step and N-step versions of the UTC. Using a Lyapunov-based analysis, we prove that the states and inputs converge to a bounded ball around the origin, whose radius depends on the bound on the nonlinearity. Using examples of a fighter jet model and a quadcopter, we demonstrate that the UTC achieves satisfactory regulation and tracking performance on these nonlinear models.

Keywords: Linear Control Systems, Mechatronic Systems, Modelling, Identification and Signal Processing
Abstract: In this paper, we exclusively consider the simple case of a 2nd order real square matrix A and define a new set of indices labeled as Non-Linear Algebraic (NLA) eigenvalues in contrast to the current literature Linear Algebraic (LA) eigenvalues. In this new NLA viewpoint, we highlight the importance of Commutativity Condition (CC) satisfaction, which is ignored by current literature linear algebra, matrix theory textbook statements and theorems. Under this new NLA viewpoint, the current literature linear algebra concepts such as Rank, Condition Number and the Singular Value Decomposition are not invoked (or not needed) and instead, we use new indices introduced in this paper, labeled as Non- Linear Algebraic (NLA) eigenvalues. Using these new indices, we provide counterexamples to one LA eigenvalue property stated in the current literature linear algebra and matrix theory textbooks, namely the Linearity property that λi(kA) = k ∗ λi(A). This counterexample is deemed to have significant impact on the way we assess the real state variable convergence of any real LTISS system.