Keywords: Control Design, Power and Energy Systems

Keywords: Automotive Systems, Control Design, Power and Energy Systems
Abstract: This paper investigates the use of zero-shot large language models (LLMs) for real-time control of coolant flow direction in a cell-level battery thermal management system (BTMS) for electric vehicles operating in cold climates. The underlying BTMS employs a bidirectional coolant loop and a nonlinear model predictive control (NMPC) framework to regulate branch flow rates and compressor speed. In the proposed framework, a locally deployed open-source LLM is introduced to generate the binary direction command, while NMPC is retained for continuous control optimization. A two-stage prompting strategy is designed to address heating and temperature uniformity objectives during cold-start operation. For comparison, two baseline strategies are considered, including Baseline I with a fixed flow direction and Baseline II, where the flow direction is determined by repeatedly solving two NMPC problems under opposite directions. Extensive simulations with multiple LLMs under deterministic decoding are conducted. Compared to Baseline II, which evaluates two NMPC problems for flow direction selection, the LLM-guided strategy removes repeated direction optimization, while achieving improved temperature tracking and uniformity. The results provide empirical insights into LLM-guided switching behaviors and their impact on thermal performance.

Keywords: Intelligent Autonomous Vehicles
Abstract: Simulation-based validation is essential for autonomous vehicles, yet generating scenarios that are simultaneously safety-critical, physically realistic, and relevant to the ego vehicle under test remains challenging. Large Language Model (LLM) provide a powerful tool for solving this problem, but existing approaches often generate scenarios from high-level textual descriptions or rely on simulators to implicitly handle feasibility, limiting trajectory-level control and grounding in real-world interactions. This paper proposes an LLM-driven framework that perturbs naturalistic interaction trajectories extracted from the nuPlan dataset to create adversarial test scenarios while preserving kinematic realism. An adversarial target identification stage ensures perturbations challenge the ego vehicle. The LLM analyzes interaction geometry and ego controller parameters to produce an interpretable multi-phase attack plan, executed by a kinematic-aware module enforcing dynamic feasibility through bounded accelerations, yawrate constraints, and friction-circle limits. Evaluation in MetaDrive under closed-loop control across diverse interaction topologies demonstrates a 23.0% attack success rate on safe-baseline scenarios. The LLM further returns a structured rationale for each attack, documenting which controller limit the perturbation targets.

Keywords: Intelligent Autonomous Vehicles, Modeling and Validation, Adaptive and Learning Systems
Abstract: Simulation-based testing has become a critical component of autonomous vehicle (AV) safety validation, particularly for rare and safety-critical interactions involving pedestrians. However, generating pedestrian behaviors that are both behaviorally plausible and safety-critical remains difficult. Purely rule-based pedestrian models often lack behavioral diversity, while purely adversarial learning methods may exploit simulator loopholes and produce unrealistic motions. To address this issue, a staged learning framework is proposed for adversarial pedestrian policy learning. The framework decouples behavioral plausibility from safety criticality by first initializing the pedestrian policy through behavior cloning using crossing demonstrations collected in CARLA and subsequently fine-tuning the policy with Twin Delayed Deep Deterministic Policy Gradient (TD3) in the same continuous control space. The proposed method is evaluated at an urban intersection in CARLA, where a controllable pedestrian interacts with surrounding traffic during crossing. Comparative experiments among BC-only, TD3 trained from scratch, and BC-initialized TD3 show that reinforcement learning substantially increases interaction criticality. BC-initialized TD3 achieves the highest collision rate and the smallest average minimum vehicle distance while exhibiting a more concentrated pedestrian speed distribution than Pure TD3. These results support staged pedestrian policy learning as an effective approach for generating challenging and task-consistent scenarios for AV testing.

Keywords: Intelligent Autonomous Vehicles, Modeling and Validation, Transportation Systems
Abstract: Crash rate is the dominant metric for evaluating autonomous vehicle safety in simulation, but it treats all collisions as equivalent and ignores differences in severity. We introduce a severity-aware evaluation framework that models safety as a combination of collision frequency and outcome severity using ISO 26262 injury levels (S0–S3), with severity estimated by calibrating simulated collisions against real-world NHTSA crash data. Evaluating autonomous driving agents trained under four scenario-generation paradigms—ChatScene, Adversarial Trajectory (AT), Adversarial Simulation (AS), and random perturbation—across 800 CARLA scenarios each, we find that crash rate and severity can lead to different interpretations of performance. The agent trained with ChatScene achieves the lowest crash rate (2.5%), but all failures are classified as S3, while the agent trained with random perturbation exhibits a higher crash rate (45.6%) yet a severity distribution closer to real-world data. These results show that crash rate alone provides an incomplete characterization of safety and that incorporating severity enables a more informative evaluation of autonomous driving systems.

Keywords: Intelligent Autonomous Vehicles, Transportation Systems, Machine Learning in modeling, estimation, and control
Abstract: Long-horizon traffic prediction is critical for automated vehicles control near intersections and queue-forming road segments, enabling energy-efficient and congestion-aware planning. This work proposes a control-oriented representation of future longitudinal motion constraints as a virtual leader acting on the ego vehicle, and develops a hybrid framework for virtual-leader prediction. In the short term, a deep learning model predicts future occupancy and flow, from which a most likely collision-free virtual leader constraint is inferred. In the long term, a queueing-theory-based model extends this constraint based on the traffic stage estimated from the short-term prediction. On the Waymo Motion Dataset, the short-term predictor achieves an area under the curve (AUC) of 0.755 for occupancy prediction and an endpoint error (EPE) of 5.862 for flow prediction, using 4s of historical data and a 5s prediction horizon. For full horizon virtual leader prediction, the proposed framework reduces the average displacement error (ADE) and final displacement error (FDE) by 18.7% and 38.4%, respectively, compared with a queueing-theory-only baseline. These results support the proposed approach as a promising framework for control-oriented virtual-leader-based motion constraint representation in complex traffic environments.

Keywords: Transportation Systems, Intelligent Autonomous Vehicles, Machine Learning in modeling, estimation, and control
Abstract: This paper introduces SociDriving, a framework for socially compliant automated vehicle (AV) control that stabilizes mixed-traffic flow and reduces congestion. Unlike adaptive cruise control (ACC), which prioritizes individual performance and often worsens congestion, SociDriving uses LSTM networks with dynamic Social Value Orientation (SVO): a single AV adapts to surrounding vehicle states (spacing, relative speeds) via a utility-based model balancing individual and collective objectives, with phi(t) shifting from self-interest to collective benefit to mitigate traffic waves. A preliminary case study using the Arizona Ring Experiments Dataset (ARED), based on a single simulation run, shows that SociDriving exhibits greater speed consistency and lower spacing variability compared to a naturalistic IDM baseline calibrated on real-world data. These initial results suggest the framework's potential to optimize mixed-traffic flow, motivating further validation with repeated trials, statistical analysis, and comparisons against fixed-SVO and standard longitudinal-control baselines. The implementation code is available at https://github.com/lilab2024/SociDriving.

Keywords: Control Applications, Control Design
Abstract: Water treatment facilities face growing demand driven by population growth and expanding agricultural and industrial activities. Energy is a major cost driver in this sector, accounting for approximately 40% of the municipal energy use in the United States. Nevertheless, most water treatment systems operate without explicitly considering time-varying water demand or the greenhouse gas (GHG) emissions associated with energy consumption. This study introduces a custom-built, bench-scale water treatment testbed and a companion digital twin for investigating emissions-aware model predictive control (MPC) under realistic sensing conditions. The test bed comprises four interconnected process tanks equipped with ultrasonic level sensors, pumps, valve actuators, and an ESP32 microcontroller within a real-time data-acquisition and control stack. The ESP32 directly receives information from the pumps and flow meters and is relayed the ultrasonic sensor data by an Arduino UNO via Universal Asynchronous Receiver Transmitter (UART). The ESP32 is integrated with a MATLAB interface for USB serial communication. A control-oriented hydraulic model serves as the digital twin and provides the prediction model for the MPC. The proposed MPC framework determines control actions that minimize energy-related GHG emissions while satisfying water-demand requirements and operational constraints. Machine-learning-based sensor correction and Kalman-filter-based state estimation are integrated to mitigate measurement errors and improve state estimates. In simulated operating scenarios, the proposed MPC strategy reduced total GHG emissions by 12% relative to a baseline controller while maintaining the required water supply. These results establish a sensor-aware digital-twin platform for closed-loop experimental validation of emissions-aware water treatment control.

Keywords: Machine Learning in modeling, estimation, and control, Robotics, Sensors and Actuators
Abstract: Accurate perception of industrial environments is essential for intelligent manufacturing, as robotic systems must reliably recognize objects and estimate their positions and orientations for downstream manipulation and automation tasks. However, existing scene reconstruction methods are commonly developed for general-purpose environments with abundant training data and relatively moderate accuracy requirements. Their performance can therefore degrade in manufacturing applications, where object-level precision is critical and task-specific data are often limited. This presentation introduces a learning-based Scene Point-cloud Registration framework for automatic industrial scene reconstruction, referred to as iLSPR. The framework combines three components: a Multi-Feature Robust Point Matching Network (MF-RPMN), which integrates raw point information and learned geometric features for accurate point-cloud alignment; a Geometric-Primitive-based Data Generation method, which efficiently generates synthetic training data; and a digital model library containing high-fidelity models of industrial target objects. During operation, point clouds captured by vision sensors are matched and registered with the corresponding object models using the pretrained MF-RPMN. To evaluate the framework, we develop the Industrial Scene Object Point-cloud Registration dataset in Isaac Sim and compare iLSPR with existing registration methods. The results demonstrate improved registration accuracy and robustness under limited-data conditions. Experiments on a real robotic manufacturing system further show that the proposed framework can reliably reconstruct industrial scenes and provide accurate spatial information for robotic operations.

Keywords: Manufacturing Systems, Control Applications, Modelling, Identification and Signal Processing
Abstract: Electrohydrodynamic (EHD) jet printing is a high-resolution additive manufacturing process whose performance depends on both stable droplet generation and accurate control of deposited feature geometry. This presentation examines two complementary studies that address these challenges at different levels of the printing process using feedforward and feedback control. The first study uses a reduced-order, physics-based model of meniscus relaxation to design modified voltage waveforms. At higher jetting frequencies, the meniscus has less time to recover between events, leading to systematic changes in droplet volume. The optimized waveform reshapes the relaxation phase to drive the meniscus toward a more repeatable state before the next jetting event. Experiments across multiple operating frequencies show reduced frequency-dependent variation in deposited droplet size compared with a conventional fixed-bias waveform. The second study uses optical density extracted from in-situ camera images to characterize deposited material during printing. For a conductive nanoparticle ink, optical density is related to deposited volume, cross-sectional area, and electrical resistivity, providing information about both geometric and functional properties. In a separate experiment using a polymer ink, real-time image-based width measurements are incorporated into a feedforward-feedback controller to regulate printed line width and track changing geometric references. Together, these studies demonstrate complementary strategies for improving robustness in EHD jet printing. Physics-based waveform design addresses repeatable dynamics of jet formation, while optical sensing and feedback enable monitoring and regulation of deposited features. Their future experimental integration could provide a more complete multilevel control framework for micro-additive manufacturing.

Keywords: Manufacturing Systems, Control Applications, Optimal Control
Abstract: Thermal history governs defect formation and coating integrity in directed energy deposition (DED), yet the temperature field entering the deposition step is seldom treated as a controlled variable. Steep gradients at the onset of coating promote cracking and residual stress and degrade interlayer bonding, while external heaters that mitigate them add cost and complexity to the machine. This work casts pre-deposition heating as a constrained feedforward control problem: given a desired axial temperature distribution at the instant coating begins, synthesize the command schedule that regulates the workpiece to it, using the deposition laser itself as the actuator so that no auxiliary hardware is introduced. A control-oriented lumped finite-volume model of a rotating thin-walled workpiece captures axial conduction, distributed surface losses, and a position-dependent heat input; the grouped coefficients are identified from infrared temperature histories, and an independent trial confirms predictive accuracy across both the heating transient and the subsequent cooling interval. Recasting the multi-pass response as an input-output convolution exposes closed-form sensitivities of the controlled output to every decision variable, including the scan-pass count, which is relaxed to a continuous quantity so that a partially executed final pass becomes admissible and total heat input can be trimmed finely. A Gauss-Newton SQP scheme then solves for pass count together with per-segment laser power and traverse speed subject to actuator bounds and command slew-rate limits, so the synthesized control inputs are directly executable on the machine. Experiments on a hybrid DED platform validate the controller, tracking flat, rising, falling, and peaked temperature references with high accuracy, whereas an uncontrolled single maximum-power pass falls well short of the elevated targets. The closed-form sensitivities further deliver order-of-magnitude reductions in solution time relative to numerical-gradient alternatives, making the formulation practical for online schedule synthesis.

Keywords: Manufacturing Systems, Machine Learning in modeling, estimation, and control, Adaptive and Learning Systems
Abstract: Milling chatter arises from regenerative tool-workpiece interaction and can cause increased cutting force, excessive tool vibration, poor surface finish, and reduced machining accuracy. Dynamic modeling is therefore important for understanding milling stability behavior. While force, acoustic, and vibration signals are commonly used for chatter detection, they provide indirect information about the tool deformation. Since direct end-mill displacement measurement is difficult during milling, an online framework is needed to infer tool deformation from measurable signals and use it for adaptive control.

This work presents a closed-loop digital twin framework for online milling stability monitoring and chatter mitigation. As illustrated in Figure, the physical CNC system continuously provides cutting force signals and machining parameters to the virtual model. The virtual model combines milling dynamics, stability analysis, and FEA-based deformation modeling with physics-informed machine learning to reconstruct the instantaneous tool vibration response. To satisfy real-time requirements, the high-fidelity FEA computation is replaced by a lightweight surrogate model that predicts the deformation response with low latency, in line with recent developments in real-time AI-driven milling digital twins. The estimated deformation field is rendered as a digital replica of the end-mill and serves as an online measure of the current stability condition.

The reconstructed tool state is further used to close the loop from the virtual model back to the physical machining process. Once the digital twin identifies vibration related to chatter, the controller modifies the machining parameters, particularly the spindle speed, and updates the CNC operation. Through this bidirectional information flow, the physical process and the virtual stability evaluation remain synchronized during cutting. Unlike conventional chatter monitoring methods that mainly depend on empirical signal thresholds or precomputed stability boundaries, the proposed framework uses the estimated tool deformation as a physics-related feedback variable for online decision making.

Keywords: Manufacturing Systems, Mechatronic Systems
Abstract: The placement of a workpiece relative to a robotic manipulator is important for ensuring that the robot tool can reach the required workpiece surfaces while satisfying task specific constraints. Typically, the placement optimization can require approximately 20-30 minutes.This computation time limits the use of existing methods when the workpiece geometry becomes available shortly before task execution. For example, in additive manufacturing and material removal processes, the geometry may change as material is added or removed. Similarly, the geometry may differ from the nominal CAD model because of process variability or may only become available after sensing the physical workpiece. This work proposes a data-driven approach for efficiently determining the workpiece pose with the highest task reachability. A robotic simulator (inverse kinematics and collision checking) and a library of workpieces are used to generate reachability data while accounting for the constraints. The generated data is then used to train a (differentiable) surrogate reachability function. During placement optimization, the learned surrogate replaces repeated inverse-kinematics and collision-checking evaluations by doing gradient descent on the surrogate function, allowing the workpiece pose to be optimized efficiently, to about 10 seconds.

Keywords: Aerospace, Control Applications
Abstract: This paper considers the problem of simultaneously controlling the impact time and impact angle of an interceptor using only lateral acceleration. In the terminal engagement phase, interceptors generally do not possess direct control over radial acceleration and are therefore restricted to maneuvering through lateral acceleration, which is the natural direction of lift. However, enhanced terminal performance often necessitates the simultaneous satisfaction of multiple terminal constraints. To address this challenge, a hierarchical sliding-mode-based guidance law is developed for the concurrent regulation of impact time and impact angle. Specifically, the proposed approach constructs a two-layer sliding architecture: the first layer comprises two sub-sliding surfaces corresponding to the impact-time and impact-angle error dynamics, while the second layer introduces a composite sliding manifold that combines the two sub-surfaces. In addition, an adaptive gain is assigned to the impact-time sub-surface to improve terminal accuracy. Numerical simulation results are provided to verify the effectiveness of the proposed guidance strategy.

Keywords: Aerospace, Control Applications
Abstract: This paper presents a nonlinear guidance strategy for a non-holonomic attacker to intercept a stationary target while avoiding being intercepted by multiple defenders. Unlike prior methods that rely on explicit knowledge of defender strategies, we quantify defender-induced threat geometrically through engagement zones, outside which the attacker can never be neutralized by the defenders. In our approach, the attacker continuously pursues the target under nominal conditions and engages evasive maneuvers only when a defender-induced engagement boundary is approached, thereby ensuring survivability. This is achieved by blending target-interception and engagement zone avoidance behaviors through a switching-based guidance law that prioritizes safety near threats while preserving nominal interception-bound trajectory elsewhere. We further show that avoiding the engagement zone corresponding to the defender nearest to the attacker is sufficient to ensure that all engagement zones are avoided, enabling scalable and computationally efficient threat management. The proposed guidance law relies solely on relative state information, requiring no global position data or defender intent estimation, making it applicable in GPS-denied and communication-constrained environments. A Lyapunov-based analysis establishes convergence to the target and ensures forward invariance of the safe set outside all defender engagement zones, while simulation results demonstrate the effectiveness of the strategy in achieving mission objectives under adversarial conditions.

Keywords: Aerospace, Control Applications, Linear Control Systems
Abstract: This paper presents a method based on Hamilton-Jacobi (HJ) reachability in a Run Time Assurance (RTA) framework to ensure the safe, collision-free operation of a deputy satellite in proximity of a chief. HJ reachability is used to define safety constraints and synthesize a backup controller that guarantees system safety even when the primary controller violates those constraints. The resulting safe sets are represented geometrically and act as a real-time safety filter. The effectiveness of the proposed method is first evaluated in simulation, where an unsafe primary controller, designed to violate the safety constraints, is used, and then validated on an actual microquadcopter platform in the Advanced Learning, Artificial Intelligence and Control laboratory at the University of Houston, serving as a surrogate for the satellite dynamics.

Keywords: Aerospace, Intelligent Autonomous Vehicles, Unmanned Ground and Aerial Vehicles
Abstract: A pre-departure scheduling methodology is developed for collision avoidance between Personal Air Vehicles (PAVs) navigating straight, intersecting corridors in an Urban Air Mobility (UAM) network. The problem is formulated as determining the minimum takeoff delay to assign to a lower-priority vehicle, such that spatiotemporal deconfliction with a higher-priority vehicle is guaranteed throughout the concurrent flight. An analytical solution is derived by employing the geometry of the corridor intersection and the relative motion of the two vehicles, yielding a set of verifiable conditions that determine whether a conflict exists and, when one does, the minimum corrective delay is expressed in closed form directly. The methodology avoids numerical optimization or iterative search, and it is computationally lightweight, thereby providing an analytical foundation for real-time scheduling in automated UAM traffic management systems. The approach is validated through numerical simulations.

Keywords: Education for modeling, estimation and control, Linear Control Systems, Aerospace
Abstract: An educational experiment using a MIMO plant has been described. This article addresses the scarcity of experimental modules that demonstrate system identification techniques using modern computational tools. It reports on a complete experience, including the use of experimental test data to derive mathematical models and their application in designing closed-loop controllers. Novel features of the system include the use of a MIMO plant (two-input, two-output) and a unique non-holonomic physical system that illustrates the challenges of modeling and controlling systems with non-holonomic constraints.

The proposed approach builds on the prior experience of the authors in designing propeller-driven mechatronic systems. This laboratory relies on the Control Systems and System Identification toolboxes in MATLAB to provide students at different training levels with laboratory experience. Both graduate and undergraduate students utilized the apparatus and completed the design tasks.

Keywords: Robotics, Aerospace, Control Applications
Abstract: In-space servicing has been proposed as a means to expand the capability and flexibility of orbital assets. In this new paradigm, repair, replenishment, upgrade, and re-purposing are all available to extend the usefulness of a satellite. Robotics, specifically serial robot arms, have been explored as a facilitator. In situations where the client does not have a prepared docking port, other means must be used for servicer-client attachment. Using multiple arms is a logical solution, with one (or more) arm(s) being used to hold onto the client and one (or more) arm(s) being used t0 service it. Additionally, servicing operations might require several disparate sequential actions. These actions might require different tooling solutions. In a scenario where the number of tools outnumbers the amount of robotic arms, a tool changer mechanism is required as the end effector. This paper showcases a real-time demonstration of this situation in a dynamically representative environment, the Robotic Orbital Control (ROC) lab. The lab houses an air bearing table which serves as a testbed for frictionless contact operations such as in-space docking or robotic action. Two heptagonal float craft represent satellite buses, which can be equipped with robotic arms and/or docking adapters. A Robot Operating System 2 (ROS2) based software architecture underlies a layered and modular autonomy and control system.

Keywords: Adaptive and Learning Systems, Control Applications, Uncertain Systems and Robust Control
Abstract: Real-time model-based control of high-dimensional nonlinear systems is computationally demanding. Conventional reduced-order model (ROM) control relies on expert tuning and offers no mechanism for online supervised reconstruction. We introduce AURORA (Autonomous Updating of ROM and Controller via Recursive Adaptation), a supervisory framework that automates ROM-based controller design and augments it with diagnostic-triggered structural adaptation. Five specialized agents collaborate through iterative generate-judge-revise cycles, while an Evaluation Agent classifies performance degradation into three operationally distinct categories (subspace inadequacy, parametric drift, and control inadequacy) and routes corrective action to the responsible agent. For linear systems with orthogonal-projection ROMs, slowly drifting parameters, and bounded measurement noise, we prove that this classification is correct, that the supervisory switching cycle preserves exponential stability subject to a dwell-time condition, and that an explicit admissible drift bound covers the redesign latency. These guarantees are stated for the linear setting; for nonlinear systems we validate the same classification empirically and outline a multiple-Lyapunov-function route toward certificates. Experiments on eight benchmark systems with state dimensions up to N = 5177 compare AURORA against expert-tuned baselines, gain-scheduled control, and online recursive-least-squares (RLS) adaptive alternatives. Controlled fault-injection experiments confirm 91% diagnostic routing accuracy. With the strongest backbone (GPT-5), AURORA achieves 6 to 12% tracking improvement over expert baselines and 4 to 5% over classical adaptive alternatives; these gains are backbone-dependent, and weaker language models can fall below the expert baseline.

Keywords: Adaptive and Learning Systems, Nonlinear Control Systems, Estimation
Abstract: This paper proposes an adaptive neural output-feedback control scheme that guarantees both safety and tracking for uncertain nonlinear systems with high relative degree. A high-gain observer and two adaptive radial basis function neural networks (RBFNNs) are developed for joint state estimation and dynamics learning, where the incorporation of RBFNNs enables reduced observer gains to mitigate the peaking phenomenon. A baseline controller is then constructed using filtered tracking error dynamics to ensure tracking performance, while safety is enforced through an exponential control barrier function-based quadratic program that provides a minimal correction. Comparative simulations verify the effectiveness of the proposed framework.

Keywords: Adaptive and Learning Systems, Robotics, Nonlinear Control Systems
Abstract: In this paper, a constrained shape servoing controller is presented for deformable objects where a subset of the state vector representing the object configuration is required to remain within constraints described by nonlinear algebraic equations. The object's deformation behavior described by a Jacobian is approximated using Fourier basis functions, whose parameters are estimated online using an Integral Concurrent learning (ICL)-based method along with the model inversion-based controller. An integral Barrier Lyapunov Function (iBLF) method is integrated with a deformation Jacobian-based adaptive controller to ensure that the selected states remain within constraints described by nonlinear functions while regulating the object shape to a desired configuration. The proposed constrained shape servoing controller enables manipulation tasks that require shape manipulation of deformable objects within arbitrary constrained spaces. Lyapunov stability analysis is performed to prove boundedness and the closed-loop stability of error systems. Simulation are performed to validate the proposed controller on a 2D deformable object, demonstrating its effectiveness in a complex shaping task subject to nonlinear state constraints.

Keywords: Assistive and Rehabilitation Robotics, Control Applications, Adaptive and Learning Systems
Abstract: Neurological injuries such as stroke or spinal cord injuries may cause damage to corticospinal pathways affecting muscle recruitment patterns. This damage and resulting movement impairments may be addressed through rehabilitation, but both the standard of care and state of the art robotic rehabilitation interventions have not yet reached their full potential. Hybrid control methods, combining exoskeleton motor control with functional electrical stimulation, provide a promising avenue to increase repetitions and engagement with the neuromuscular system. However, hybrid control systems produce complex, highly non-linear dynamics that may limit real-world applications. To accommodate for unmodeled dynamics for robust control methods, neural networks learn in real-time to reduce model uncertainty. Within these error minimizing control methods, there are design choices which have underexplored trade offs in the suitability of movement therapies, such as smoothness. In this paper, we implement multiple real-time neural network based hybrid controllers for a bi-directional 2-DoF trajectory tracking task. Furthermore, we compare human and control performance for three control schemes for real time model adaptation and the implications for extended use in rehabilitation interventions. Additionally, we investigate activation functions and their impact on control performance and crucially, the impacts on movement smoothness. Results from simulation and experiments with three healthy participants indicate that activation functions, such as tanh, and controllers, such as L-GraD+RMS, have comparable error performance to similar controllers. Further, these architectures generated smoother trajectories, key to movement quality objectives for hybrid exoskeleton rehabilitation interventions.

Keywords: Estimation, Adaptive and Learning Systems, Stochastic Systems
Abstract: This paper presents a method for creating a real-time Digital-Twin of a warehouse with information about moving objects using multiple camera views. In industrial settings, tracking of mobile platforms is traditionally carried out using technologies such as visual or radio tags, indoor positioning systems, Ultra-Wide Band, or WiFi, all of which require devices in the environment or on vehicles. Cameras, though ubiquitous, are challenging to use because of issues with synchronization, calibration, occlusions, and maintaining unique track IDs across multiple targets and cameras. We propose a method that combines information from a network of cameras to create a capability to track objects or vehicles in large environments. Our approach uses YOLO for detection and a special fusion algorithm to merge multi-camera observations into a unified global view. This process removes ghost duplicates and creates a single accurate observation for every vehicle, on a bird's eye view map. A Kalman Filter to predict movement during occlusions, ensuring continuous tracking. We tested the system on two specific scenarios. The first is a blackout test that measures how accurately the math guesses a mobile objects path during occlusions. The second is a crossing test that ensures multiple objects can cross paths perpendicularly without the system swapping their identities. The results show high localization accuracy and reliable identity retention, providing a solid foundation for warehouse safety and traffic management systems.

Keywords: Robotics, Estimation, Adaptive and Learning Systems
Abstract: This paper presents a three-dimensional position estimation system for centimeter-scale robotic platforms using magnetic field sensing. The system employs an actively controlled permanent magnet to generate directional magnetic field variations, and a compact array of onboard magnetic sensors to capture spatial measurements. A learning-based magnetic field model is developed to represent the nonlinear relationship between robot pose and magnetic field observations. This model is integrated within an Unscented Kalman Filter to jointly estimate the robot’s three-dimensional position and yaw angle. Active orientation of the external magnet is utilized to improve observability and reduce ambiguity in the magnetic field measurements. Experimental validation on a three-dimensional translation–rotation stage demonstrates that the proposed system achieves high accuracy and robustness over an extended workspace, outperforming passive magnetic localization approaches.

Keywords: Machine Learning in modeling, estimation, and control, Power and Energy Systems, Automotive Systems
Abstract: Achieving extreme fast charging in li-ion batteries requires models that can accurately predict lithium plating and its impact on capacity degradation. Plating introduces both loss of cyclable lithium and loss of active material, as dendritic growth reduces the available surface area for intercalation reactions. Most degradation models predict the plating onset and stripping behavior as side reactions, however fail to capture the morphological effects caused by dendritic growth. This paper introduces a morphology-informed surface blockage model for representing the loss of active material induced by plating. The blockage metric is defined starting from two-dimensional phase-field simulations and represented by a bounded analytical surrogate obtained using physics-constrained symbolic regression. Integration with an established plating/stripping model illustrates the coupled evolution of loss of cyclable lithium and loss of active material during a single fast-charge/rest event.

Keywords: Machine Learning in modeling, estimation, and control, Power and Energy Systems, Modelling, Identification and Signal Processing
Abstract: This paper develops a machine learning-based simulation model of lithium-sulfur (Li-S) battery physics. Li-S batteries are known for their attractive energy densities. Moreover, a significant literature exists on modeling, identifying, estimating, and controlling their dynamics. Most models in this literature represent Li-S battery physics using complex, nonlinear, and numerically stiff differential algebraic equations (DAEs). The main contribution of this paper is to develop a backpropagation neural network that approximates the discrete-time solution of these physics-based DAE models for different battery cycles. This addresses the two most critical numerical challenges associated with physics-based Li-S battery models, namely, their implicit nature and numerical stiffness. Simulation studies show that the resulting neural network captures Li-S battery dynamics with very high accuracy over a broad range of discharge rates.

Keywords: Nonlinear Control Systems, Control Design, Power and Energy Systems
Abstract: Port-Hamiltonian (pH) systems provide a physically consistent framework for modeling nonlinear multi-energy systems by explicitly capturing energy storage, interconnection, and dissipation. Their structure naturally supports stability analysis, passivity-based control, and modular system representation. However, directly employing nonlinear pH models in constrained nonlinear model predictive control (NMPC) poses a significant computational challenge, particularly in real-time applications where constrained optimization must be solved repeatedly at high sampling rates. While linear parameter-varying (LPV) and quasi-LPV (qLPV) representations have been widely adopted to improve computational tractability, most existing embedding methods do not explicitly preserve pH structure. This limits physical transparency and can reduce consistency between model interpretation and controller design. This work addresses this gap by developing a structure-preserving qLPV embedding framework for nonlinear pH systems that retains the underlying energy-based representation while enabling efficient predictive control.

Keywords: Optimal Control, Power and Energy Systems, Robotics
Abstract: Energy optimization is a central concern across many engineering fields. Optimal control theory provides a powerful tool for optimizing prescribed objective functions. Recently, there has been increasing interest in adopting a whole-system perspective in optimization, rather than considering individual subsystems in isolation. This shift is motivated by the complexity of modern multiphysics, interconnected systems that necessitate the formulation of more comprehensive efficiency objectives. In this work, a thermodynamics-based objective function is defined and optimized for an electrically actuated double pendulum used as an energy harvester. The approach is inspired by the second law of thermodynamics (SLT), enabling the formulation of an objective function based on a generalization of the concepts of lost work and entropy generationfrom the Gouy–Stodola theorem. Numerical simulations are performed to minimize the proposed objective and the results are compared with those obtained from direct energy maximization. The results show excellent agreement, highlighting the potential of this approach for large-scale, highly complexsystems.

Keywords: Power and Energy Systems, Machine Learning in modeling, estimation, and control
Abstract: Battery health management is heavily dependent on reliable State-of-Health (SoH) estimation to ensure battery safety with maximized energy utilization. Although online state-of-health (SoH) estimation can effectively track battery degradation, it requires continuous battery data acquisition. In addition, model-based SoH estimation methods rely on accurate battery model knowledge, whereas data-driven approaches often suffer from limited interpretability. In contrast, analytical characterization of SoH will offer a direct and tractable handle on battery performance degradation, while also establishing a foundation for further analytical studies toward effective battery health management. Thus, in this work, we propose a Kolmogorov–Arnold Network (KAN)-based data-driven pipeline to establish a functional relationship for SoH degradation using battery temperature data. Specifically, we learn long-term battery thermal dynamics and battery heat generation via learnable activation functions of our KAN model. We utilize the learned mapping to obtain an explicit functional relationship between SoH degradation and cycle number. The proposed pipeline was validated using real-world data, yielding a closed-form analytical formula of SoH degradation with high accuracy.

Keywords: Power and Energy Systems, Unmanned Ground and Aerial Vehicles
Abstract: While motion control for small autonomous vehicles receives significant attention, less emphasis has been placed on the coupling between motion and energy. Coordinating this dynamic coupling across the mechanical, electrical, and thermal domains is both mission-critical and safety-critical. To facilitate advancements in this area, this paper proposes a laboratory scale testbed for energy-aware autonomy. The testbed pairs a small autonomous ground vehicle with a temperature-sensitive auxiliary payload. In addition to the design, modeling, and open-loop experimental validation of the proposed platform, the paper also presents a closed-loop simulation study with model predictive control, exemplifying how the platform can be used to advance frameworks for energy-aware autonomy.

Keywords: Estimation, Multi-agent and Networked Systems, Discrete Event Dynamic Systems
Abstract: This paper studies distributed adaptive state estimation for discrete-time nonlinear systems with unknown source dynamics over directed communication networks. Each sensing agent estimates the source state using only local measurements and information exchanged with neighboring agents, enabling a fully distributed implementation without requiring shared excitation or control inputs. A normalized adaptive estimation scheme is proposed to identify unknown linear and nonlinear dynamics while ensuring robust discrete-time adaptation. A Lyapunov-based analysis establishes input-to-state stability (ISS) of the estimation error dynamics, guaranteeing bounded adaptive parameters under bounded disturbances and asymptotic convergence of the estimation errors in the disturbance-free case under suitable conditions. To characterize the network-induced coupling, explicit norm-based and LMI-based Schur stability conditions are developed for the coupling operator, including a robust formulation accounting for bounded model uncertainty. Numerical simulations on star, cyclic, and path communication topologies demonstrate accurate distributed state estimation and validate the proposed stability conditions. Computational results further show that the proposed estimator scales efficiently with the network size.

Keywords: Estimation, Multi-agent and Networked Systems, Modeling and Validation
Abstract: This paper presents a framework for 3D pose tracking of humans in multi-person environment via identity association. By combining Faster-VoxelPose with a kinematic model in a Kalman Filter, we leverage spatio-temporal state prediction to resolve the multi-object data association problem. Unlike recent approaches that rely on deep-learning based models integrated with the pose estimator, we employ a probabilistic matching strategy utilizing Mahalanobis distance and Hungarian algorithm that does not rely on a specific pose estimator. Further, we introduce a limb length metric that enforces biometric consistency of each tracked pose on the data matching. We validate the proposed approach on the benchmark CMU Panoptic dataset as well as custom real-world laboratory sequences. Quantitative evaluation using Multi-Object Tracking Accuracy (MOTA) and ID F1 Score (IDF1) metrics demonstrates that our framework achieves robust ID tracking with minimal identity switches.

Keywords: Estimation, Robotics, Multi-agent and Networked Systems
Abstract: Lane detection and mapping has become crucial for many vehicular applications, including lane departure warning, localization, adaptive cruise control, as well as level 4 autonomous driving. Existing approaches strive to balance the precision of lane modeling and its onboard execution footprint. In this paper, we present a fully distributed localization and lane mapping framework using visual-inertial (VI) and global navigation satellite system (GNSS) measurements, where each vehicle utilizes a loosely-coupled GNSS-VI smoothing algorithm by gathering measurement constraints from its onboard sensors as well as those of neighbors. Besides, we adopt an innovative road-map model by integrating the B'ezier curve and an adaptive keypoints combination algorithm. This parameterized map allows us to reduce a minimum of 30% generalized keypoints among a 10km highway sampling, which expedites the estimation while maintaining nearly equivalent mapping accuracy. The proposed framework is validated extensively in Monte-Carlo simulations. The results demonstrate that our cooperative framework yields improvements of at least 20% in localization and 12% in mapping compared to the individual GNSS-VI smoothing.

Keywords: Linear Control Systems, Multi-agent and Networked Systems, Control Applications
Abstract: Fragility measures how far a dynamical network is from instability and is important in characterizing the robustness of dynamical networks. Recent work showed that the fragility of second-order dynamical oscillator networks is related to localized eigenvectors. In this work, we compare the effects of network structure on fragility due to structured perturbations targeting localized and delocalized eigenvectors in dynamical oscillator networks for Erdős–Rényi, Barabási–Albert, and Watt-Strogatz random networks. Our results show the relationship between network structure and fragility, linking eigenvector localization to heightened sensitivity to perturbations across varying network structures irrespective of network size. Our results will inform future design and control of dynamical oscillator networks.

Keywords: Multi-agent and Networked Systems, Nonlinear Control Systems, Aerospace
Abstract: A new strategy of parametric stabilization of non-consensus equilibria in decentralized leaderless attitude cooperative control of a multi-agent rigid body system is proposed using feedback of relative attitudes in the form of rotation matrices with time-periodic gains and relative angular velocities on an undirected and connected communication graph. As an alternative to control protocols with desired relative attitudes, it is shown that parametric stabilization can provide local stability to these non-consensus equilibria. The technique is illustrated for undirected networks of two and three rigid body agents.

Keywords: Nonlinear Control Systems, Optimal Control, Multi-agent and Networked Systems
Abstract: This paper develops a consistent implicit discretization framework for optimal finite-time stabilization of nonlinear discrete-time systems. Building on the link between discrete-time finite-time stable controllers and the implicit Euler method, we construct controllers that preserve exact finite-time convergence via a coordinate transformation based on the canonical homogeneous norm, yielding a closed-form control law that eliminates numerical chattering. An optimal finite-time control problem is then formulated and sufficient Bellman conditions are derived, with the steady-state solution serving as a Lyapunov function guaranteeing both finite-time stability and optimality. A numerical example involving multi-agent finite-time consensus is provided to illustrate the framework.

Keywords: Adaptive and Learning Systems, Multi-agent and Networked Systems, Linear Control Systems
Abstract: A discounted model-free reinforcement learning framework based on policy iteration is developed for distributed optimal control of discrete-time dynamic graphical games. Each agent learns its control policy using local neighborhood information while minimizing an individual performance index and synchronizing with a leader. A discount factor is incorporated into the policy evaluation step to facilitate convergence of the learning process. Policy evaluation and policy improvement are performed directly from input state data without requiring explicit knowledge of the system dynamics. Numerical results for a four-agent directed graphical game demonstrate convergence of the learned control gains, bounded system trajectories, and asymptotic convergence of the synchronization errors. The results demonstrate the effectiveness of discounted policy iteration for distributed Nash control and leader-follower synchronization in multi-agent systems.

Keywords: Machine Learning in modeling, estimation, and control, Modelling and Control of Biomedical Systems
Abstract: Wastewater viral data offer a near real‑time, population‑level indicator of disease transmission dynamics and may allow healthcare systems to respond promptly and reduce the spread of infections. This project explores the National Wastewater Surveillance System (NWSS) data with machine learning (ML) techniques to predict the relative outbreak timing of major respiratory infectious diseases, including SARS-CoV-2, influenza and respiratory syncytial virus (RSV). By analyzing temporal and geographic correlations among the wastewater data of these viruses from 53 U.S. states and territories, this project aims to uncover the time lags between wastewater viral activities and assess the impact of demographic factors. In addition, this project leverages ML-based regression models to evaluate the feasibility of using one virus as an early indicator for others.

Results show that SARS-CoV-2, influenza and RSV virus concentrations in wastewater exhibit clear seasonal patterns and significant correlations in their onset, peak and offset time. SARS-CoV-2 signals in wastewater typically precede influenza and RSV signals by at least 30 weeks. However, the predictive power of SARS-CoV-2 signals for influenza and RSV signals varies greatly across U.S. regions and is not strongly associated with population density, age, or sex ratio.

Overall, ML models demonstrate potential for forecasting influenza and RSV data using SARS-CoV-2 wastewater data, although predictive accuracy requires further improvement. These findings highlight the promise of integrating wastewater surveillance with modeling approaches to improve outbreak prediction and public health response.

Keywords: Aerospace, Nonlinear Control Systems
Abstract: Intercept control is widely used across aerospace and autonomous systems, including homing missile guidance, spacecraft rendezvous and proximity operations, and autonomous interception of moving targets by aerial and ground robots. Intercept guidance is fundamentally a tracking problem in which an interceptor must minimize the error between its trajectory and the trajectory required to intercept a moving target. Classical proportional navigation (PN) guidance is widely used because of its simplicity and effectiveness; however, the commanded motion generated by the guidance law cannot be followed perfectly when the interceptor has non-negligible dynamics and limited maneuverability. This discrepancy introduces tracking error, which can propagate through the guidance loop and ultimately increase the terminal miss distance. A central research challenge is therefore the development of intercept strategies that minimize tracking error while accounting for the dynamic response of the interceptor, particularly for rapidly maneuvering targets and high-bandwidth engagement scenarios. The main contribution of this work is to improve tracking by adding a feed-forward signal, with model based inversion, to a traditional Adaptive Proportional Navigation (ADPN) scheme. This proposed approach reduces the RMS terminal miss distance from 9.73 m for ADPN alone to 0.565 m, corresponding to a 94.2% reduction over the target-maneuver phase sweep.

Keywords: Control Applications, Control Design, Modelling, Identification and Signal Processing
Abstract: Fast charging of lithium-ion batteries is critical for the widespread adoption of electric vehicles (EVs), but aggressive charging can accelerate degradation and introduce safety risks, particularly lithium plating (LP) and over-temperature conditions. This paper presents an enhanced lithium-plating-aware fast charging framework that integrates observability-aware parameter identification, half-cell state estimation, and Control Barrier Function (CBF)-based safety control. An observability-aware parameter identification procedure is first developed for a dual half-cell equivalent circuit model (ECM), enabling the identification of model parameters relevant to the estimation of electrode-level states. An Extended Kalman Filter (EKF) is then designed based on the identified model to estimate the anode potential, providing a model-based indicator for LP mitigation that cannot be directly measured during battery operation. The proposed estimation and control framework is evaluated under fresh, aged, and lithium-plated cell conditions using a validated, physics-based high-fidelity battery model implemented in AVL CRUISE™ M. As a baseline, a conventional power-based charging controller with constraint-activated proportional-integral (PI) regulators is implemented to enforce temperature and anode-potential safety limits. The results demonstrate that the proposed CBF-based controller provides improved charging performance while maintaining the imposed thermal and lithium-plating constraints across different cell conditions. These results highlight the potential of combining observability-aware half-cell modeling, electrode-level state estimation, and computationally efficient CBF-based control for safe and fast charging of lithium-ion batteries.

Keywords: Healthcare systems, Modelling and Control of Biomedical Systems
Abstract: Acute mental stress induces cardiovascular (CV) arousal that can impair health and performance, while current pharmacological approaches are not suited for just-in-time intervention. This study evaluates wearable-based detection and mitigation of acute stress-induced CV arousal via closed-loop transcutaneous median nerve stimulation (TMNS). We analyzed 40 experiments from 20 healthy participants undergoing stress and TMNS interventions. Six CV physio-markers capable of capturing CV arousal to acute mental stress were derived from electrocardiogram, seismocardiogram, and photoplethysmogram signals and integrated into a synthetic multimodal variable (SMV) representing overall CV arousal, which served as a feedback variable for closed-loop control. A μ-synthesis-based closed-loop controller was designed to regulate the SMV through TMNS, and a prototype wrist-worn device was developed for CV arousal detection and TMNS delivery. Results showed that SMV reliably tracked stress-induced CV arousal, which was partially mitigated by closed-loop TMNS. A pilot human study further confirmed the functionality of the prototype device in real-time detection and stimulation capability. These findings support the feasibility of wearable closed-loop TMNS for just-in-time mitigation of acute mental stress-induced CV arousal.

Keywords: Machine Learning in modeling, estimation, and control, Adaptive and Learning Systems, Control Applications
Abstract: Achieving optimal control of nonlinear process systems under changing operating conditions represents a critical challenge; machine learning approaches such as reinforcement learning (RL) have significant appeal for this purpose. This study presents a RL framework that learns effective control strategies through direct interaction with the process, eliminating the need for case-specific tuning under varying setpoint and disturbance changes. A spherical reboiler is examined as a benchmark nonlinear process system. Internal Model Control (IMC)-based Two-Degree-of-Freedom Proportional-Integral-Derivative (2DoF PID) controller tuning is used to establish a reference closed-loop response for setpoint tracking and disturbance rejection, which guides formulating a control-oriented reward function. The desired control response provides guidance for reward design, encouraging actions that respect actuator limits, avoid abrupt control moves, and maintain desired control variable performance in a manner appealing to the user. Through appropriate reward shaping and hyperparameter tuning, a Twin Delayed Deep Deterministic Policy Gradient (TD3) based RL framework achieves reliable and consistent control, successfully maintaining the desired liquid level while producing smooth outlet flow adjustments. Compared to the 2DoF IMC PID controller, the RL framework improves disturbance rejection by reducing transient deviations and overshoot, demonstrating its potential for adaptive control without demanding case-specific tuning. This work provides insights and the foundation for developing general RL frameworks for nonlinear process control with minimal user interaction. Future efforts will focus on learning under changing disturbance and setpoint conditions, with transfer learning reducing the amount of retraining required for new systems; the framework will also be extended to improving Model Predictive Control (MPC).

Keywords: Human-Machine and Human-Robot Systems, Cyber physical systems, Robotics
Abstract: Shared control blends human and autonomous commands to improve performance in complex human-robot tasks. Existing studies often represent shared-control level using a small number of discrete conditions that change the operator interface, command abstraction, or allocation of responsibilities between human and autonomy. These coupled changes make it difficult to isolate the effects of control-authority allocation on work-load and task performance. Continuous shared control instead blends human and autonomous commands through a level-of-autonomy parameter, α, enabling fine-grained adaptation without changing the underlying interface. However, the relationships among continuous autonomy level, operator workload, and task performance remain insufficiently characterized. This work uses physiological workload indicators to advance the development of a closed-loop, workload-aware adaptive-assistance controller. In a human-subject confined-space inspection task using an end-effector-mounted camera, varying α produced measurable changes in both cognitive workload — often calculated using eye-tracking metrics — and task performance. Both exhibited non-monotonic relationships with α. The occurrence of both a performance maximizing and workload minimizing level of autonomy suggests that 1) task performance and mental workload react similarly in response to varying levels of α, and consequently that 2) α can be intentionally modulated to influence task performance during a teleoperation task using measured workload. The main contributions of the current work are the following. (i) Multimodal mental-workload estimation: Operator workload is estimated by fusing physiological and interaction signals from a Tobii eye tracker, Empatica wristband, and Force Dimension joystick. (ii) Online task-performance prediction: Future task performance is predicted in real-time from current-window workload estimates. (iii) Human-subject validation: The relationships among continuous level of autonomy, workload, and performance are experimentally characterized in an end-effector-camera inspection task in a confined space

Keywords: Linear Control Systems, Control Design, Adaptive and Learning Systems
Abstract: Safe and resilient control of autonomous systems depends on an accurate system model, even in the presence of nonlinear faults. The objective of this paper is to estimate the bounds on nonlinear fault dynamics along with bounds on the system states. The proposed Interval Gaussian Process architecture includes an interval estimator, fault detector, and regression algorithm. The regression algorithm and interval estimator are integrated to provide bounds on the fault dynamics and system states, respectively. The fault detector triggers the learning of a fault model using a Gaussian Process. Working together with a robust controller, this architecture aims to provide robust estimation and control for linear systems subjected to nonlinear fault dynamics. Finally, the performance of this architecture is demonstrated for a UAV with several faulty cases.

Keywords: Modeling and Validation, Control Design, Linear Control Systems
Abstract: The performance limits of a linear quadratic regulator designed from a hover linearization of a nonlinear quadrotor system are investigated. The nonlinear dynamics are linearized about the hover equilibrium, and an LQR controller is synthesized for three-dimensional trajectory tracking. The closed-loop system is evaluated on the nonlinear plant under increasing trajectory aggressiveness, aerodynamic effects, and parametric uncertainty in the mass and moments of inertia. Tracking performance is quantified by position root-mean-square error, maximum tracking error, and control effort. A performance envelope is constructed to characterize the operating conditions over which the hover-designed LQR maintains acceptable tracking performance and to identify the conditions for which the local linearization becomes inadequate. The results provide a quantitative assessment of the applicability and limitations of hover-based LQR control for nonlinear quadrotor trajectory tracking.

Keywords: Large Scale Complex Systems

Keywords: Control Design, Electromechanical systems, Power and Energy Systems
Abstract: Controlling a DC motor using conventional model-based controllers often fails to handle dynamic load change scenarios. To overcome such limitation, data-driven Artificial Intelligence (AI) approaches, such as Reinforcement Learning (RL), have demonstrated superior performance compared to conventional controllers. This research successfully proposes the implementation of a Soft Actor-Critic (SAC) based RL controller for a four-quadrant DC chopper to regulate DC motor speed, specifically considering dynamic load change scenarios common in the real world. The training results confirm that the algorithm successfully converged to an optimal policy. The system architecture utilizes a four-quadrant DC chopper circuit, that allows motor operation in forward and reverse motoring modes. The reward function is designed to optimize performance by penalizing speed error and high current consumption. A comparative analysis with a Self-Tuning Regulator (STR) certifies the credibility and enhanced robustness of the designed SAC controller. Results show that the SAC controller consistently achieved a smaller average speed deviation than the STR controller under higher load disturbances (e.g., 0.025 Nm and 0.05 Nm).

Keywords: Control Design, Mining, Mineral and Metal Processing
Abstract: Large-format lithium-ion batteries offer a superior alternative to traditional diesel engines and high-voltage power cables for use in underground mining operations. However, their failure through thermal runaway can pose hazardous conditions, threatening lives and infrastructure. Operations develop high-fidelity deterministic models to study their impact. To address this limitation, we develop a stochastic modeling and control framework for temperature and smoke evolution. State-space models are constructed to represent the coupled dynamics of temperature, smoke concentration, and sprinkler cooling effects across discretized tunnel nodes. The stochastic nature of the heat release rate is modeled using a regime-switching lognormal process capturing the growth, developed burning, and decay phases of the fire. A sensor-based feedback control strategy is then designed to regulate sprinkler flow commands using real-time measurements of temperature and smoke concentration. Simulation results demonstrate that the proposed approach significantly reduces both temperature and smoke propagation along the tunnel, improving safety conditions during battery fire events.

Keywords: Linear Control Systems, Control Design, Security and Privacy
Abstract: The paper presents stability conditions and stabilizing control for a linear system subject to switching based on logical decisions. The switching is due to a cyberattack, but the methodology can also handle operational or environmental changes that require changing the dynamics to maintain satisfactory operation. Several examples demonstrate the application of the stability conditions to a system subject to switching.

Keywords: Mechatronic Systems, Control Design, Robotics
Abstract: The advent of dexterous multi-fingered robotic grippers opens the door for flexible, contact-rich interaction, which ultimately requires compliant control. Compact, highly-geared actuators used in such hands present a fundamental obstacle to impedance control: their reflected task-space inertia is large enough that Jacobian-transpose-only impedance commands systematically under-realize the prescribed closed-loop dynamics. This paper presents task-space impedance control and a model-based impedance computed-torque (MBICT) control law of thumb pinching interactions using a PSYONIC Ability Hand. The MBICT includes system identification of the thumb flexor dynamics, online inertia shaping, viscous friction compensation, and a force-sensing resistor feedback loop for contact disturbance rejection. One set of experiments include a commanded step position of 5mm while pinching tweezers. MBICT control achieves the designed critically-damped response to within 1.8% or 0.09mm steady-state error and zero overshoot, while a Jacobian-transpose simple-impedance baseline at the same impedance gains reaches only within 12% of the desired commanded step position, resulting in 0.6mm steady-state error. MBICT reduces steady-state error by 6x on closing and 13x on opening, and roughly halves the transient duration. The force-sensing resistor feedback further reduces peak displacement under contact disturbance by 17%.

Keywords: Robotics, Control Design, Mechatronic Systems
Abstract: Mobile manipulators performing contact-rich tasks can realize the same Cartesian objective through different arm--base configurations, producing different directional transmission characteristics and contact responses. We augment reduced-coordinate Cartesian impedance control for a nonholonomic mobile manipulator with contact-frame directional manipulability shaping. A quadratic program prioritizes impedance tracking while incorporating secondary null-space objectives for wall-normal force transmission, tangential-plane velocity manipulability, and tangential isotropy. In a simulated layered-wall task with a soft-to-stiff transition, normal shaping increased all three considered metrics and sped up progression, whereas tangential shaping emphasized tangential manipulability and favored tangential regulation. Among the tested cases, mixed normal--tangential shaping provided the clearest practical compromise between wall-normal progression and tangential response.

Keywords: Healthcare systems, Control Design, Estimation
Abstract: This paper examines the problem of controlling the automated clinical administration of propofol, an anesthetic drug, to ensure patient safety. Existing anesthesia controllers typically adjust drug infusion rates to achieve a target patient sedation level. Our goal is to complement this previous research by developing an algorithm that (i) provides formal anesthesia safety guarantees and (ii) can be used either as a standalone controller or as a supervisory safety governor for existing controllers. Specifically, we introduce two barrier functions that quantify the safety implications of over- and under-dosing. We then construct novel nonlinear second-order barrier constraints that achieve safety in a “graceful” manner, by providing multi-layered safety guarantees, even though the dynamics of the underlying barrier functions have a relative degree of 2. Finally, we design a safety control algorithm that enforces these barrier constraints in the absence of full state measurements by combining the constraints with an interval observer. Simulation results show that the proposed approach succeeds in enabling the safe induction and maintenance of target sedation levels, with reasonable induction times and“graceful” multi-layered safety assurances.

Keywords: Automotive Systems, Estimation
Abstract: Heavy-duty articulated vehicles frequently suffer from geometric off-tracking during cornering, posing significant safety hazards that remain largely unmonitored by contemporary tractor-centric Advanced Driver Assistance Systems (ADAS). To address this blind spot, this paper presents a Lane Departure Warning (LDW) framework that combines monocular computer vision with tractor-based articulation angle estimation. The perception pipeline leverages Inverse Perspective Mapping (IPM), color-aware edge detection, and RANSAC-based joint polynomial modeling to continuously map local lane boundaries. To accurately project the trailer’s swept path relative to these boundaries, two parallel hitch angle estimators were developed and comparatively evaluated: a parameter-dependent 5-axle dynamic model using an Extended Kalman Filter (EKF) and a parameter-independent kinematic estimator. Experimen- tal validation utilizing a full-scale Class 8 Peterbilt 579 with ground truth demonstrates that while both estimators perform well in straight-line driving, the kinematic approach performs better during cornering. Furthermore, on-road testing successfully validates the system’s ability to accurately predict trailer boundary oversteps even when the tractor remains safely centered, establishing a reliable perception foundation for future automated off-tracking mitigation.

Keywords: Automotive Systems, Estimation
Abstract: Accurate state of charge (SOC) estimation is critical for the safe and efficient operation of electric vehicles (EVs). Lithium Iron Phosphate (LFP) batteries, widely adopted for their high safety and long cycle life, present a unique challenge due to their flat open-circuit voltage (OCV)–SOC profile, which limits the accuracy of voltage-based methods. This paper proposes an enhanced SOC estimation technique using Electrochemical Impedance Spectroscopy (EIS). The method exploits the monotonic relationship between SOC and impedance characteristics by fitting a geometric circular or elliptical arc to Nyquist plot data within a targeted frequency range (5–500 Hz). The extracted radius serves as a robust SOC indicator, even in voltage-flat regions. Experimental results demonstrate that the proposed approach maintains accuracy across varying temperatures and immediately after cycling, eliminating the need for prolonged relaxation periods. This method offers a practical pathway for on-board implementation in advanced battery management systems.

Keywords: Automotive Systems, Machine Learning in modeling, estimation, and control, Adaptive and Learning Systems
Abstract: Adapting vehicle control systems to low-friction conditions is safety-critical yet data-scarce, as on-surface collection poses significant risk. This paper presents a systematic, multi-seed evaluation of transfer learning for Model Predictive Control (MPC) cloning across friction coefficients, using gradient boosting (XGBoost) trained on real data from a Renault Austral prototype (>1.9M samples, 104 files). Three transfer methods, CORrelation Alignment (CORAL), Continue Training with early stopping, and weighted Ensemble Learning, are benchmarked against three baselines (scratch, hyperparameter-matched scratch, and mixed data pooling) across 5 seeds and target domain sizes of 500 10,000 samples. Our key findings are: (i) Mixed-data pooling achieves the strongest overall performance (+73.1% RMSE reduction at n=500, improving consistently to +10.9% at n=10,000), demonstrating that naive source data augmentation is a highly effective baseline that prior work has overlooked; (ii) Ensemble Learning provides consistent, reliable improvement (+3.7% to +12.4%) across all sample sizes and is never harmful, making it the recommended approach when source data cannot be directly pooled; (iii) Continue Training achieves meaningful gains only in the extreme low-data regime (n≤1,000: up to +28.2% at n=1,000) but degrades severely beyond that; (iv) CORAL fails catastrophically throughout, and (v) HpMatch matches Baseline everywhere, confirming that Continue Training’s low-data gains stem from warm-start knowledge transfer rather than from its conservative hyperparameters.

Keywords: Automotive Systems, Machine Learning in modeling, estimation, and control, Transportation Systems
Abstract: Battery Electric Vehicles (BEVs) are expected to play a central role in sustainable transportation; however, range anxiety remains a persistent barrier to adoption. Accurate energy consumption (EC) prediction (ECP) is essential to mitigate this concern, as it underpins range estimation, eco-route planning, and charging station guidance. Existing ECP algorithms are broadly categorized as physics-based and data-driven approaches. Physics-based approaches rely on vehicle dynamics and powertrain relationships, providing transparency and interpretability, but often require simplified assumptions and many parameters, while data-driven approaches learn EC patterns directly from vehicle, road, and environmental data. Driver behavior remains a critical yet underexplored factor. Aggressive driving can increase vehicle EC by over 15% relative to conservative driving. Recent ECP frameworks incorporating driver behavior often use simplified models that cannot capture real driving variability or black-box models that obscure its effect on EC. This work proposes a driver-adaptive ECP framework for BEVs using structured behavioral uncertainty. A physics-based baseline model is developed from segment-level driving features, and driver-induced variability is represented as a multiplicative correction to baseline EC. A Gaussian mixture model constructs a probabilistic latent representation of driving style, while a quantile regression model maps this information to an energy correction factor. The framework is evaluated using real-world fleet data from highway and urban driving. Benchmarked against the baseline, the proposed method reduces validation mean absolute error and root-mean-square error by 18-19% and 17-37%, respectively, with percentage error improving by 1.2-1.9 percentage points. Relative to XGBoost, the proposed method achieves comparable accuracy while retaining full interpretability, since each predicted deviation can be traced to a specific driving-style component. An ablation study further confirms that both the Gaussian mixture representation and the correction structure are necessary for prediction accuracy.

Keywords: Automotive Systems, Optimal Control, Transportation Systems
Abstract: Mobile charging stations (MCS) offer a promising solution to alleviate charging scarcity in highway corridors with limited fixed charging infrastructure. This paper studies the scheduling of an MCS under a charge-as-a-service paradigm, where each electric vehicle (EV) can be served at multiple feasible locations along its route, which leads to a coupled spatial–temporal coordination problem. To address this challenge, we develop a mixed-integer linear programming (MILP) model that jointly optimizes charging location selection, service sequencing, and timing decisions. The objective minimizes a weighted sum of EV waiting time and service duration, MCS travel cost, and onboard battery capacity cost, where service duration captures the impact of location-dependent energy demand on charging time. This formulation explicitly captures the trade-off between user inconvenience and MCS operational efficiency under flexible service locations. The proposed strategy is validated across multiple case studies, and the results demonstrate that it consistently outperforms the first-come-first-serve baseline in both scalability and overall system performance. In particular, the proposed approach achieves cost reductions ranging from 17.63% to 34.5%, with the relative savings increasing as the demand level grows.

Keywords: Intelligent Autonomous Vehicles, Automotive Systems, Transportation Systems
Abstract: Connected and automated vehicles enable energy-efficient trajectory planning, but the mechanisms through which trajectory optimization translates into energy savings can differ substantially across powertrain architectures. Existing studies typically consider individual powertrains using different models, control formulations, and operating conditions, making direct comparison difficult. This work develops a unified Koopman-based model predictive control framework for electric vehicles (EVs), internal combustion engine vehicles (ICVs), and hybrid electric vehicles (HEVs), using a common lifted-prediction structure and quadratic-programming formulation while retaining architecture-specific powertrain dynamics. Under common vehicle parameters, lead-vehicle trajectories, constraints, and evaluation conditions, the framework enables controlled comparison of both energy savings and the associated energy-flow mechanisms. Results show that trajectory smoothing reduces energy consumption across all three powertrains, but the sources of improvement differ substantially. EV savings are primarily associated with reduced braking and regenerative energy cycling, ICV savings arise largely from reduced wheel-energy demand and engine losses, and HEV savings additionally involve changes in engine operating efficiency and power split between propulsion and battery charging. These results demonstrate that energy-efficient trajectory planning should be understood not only through reductions in vehicle-level power demand, but also through how each powertrain converts, distributes, and dissipates energy internally.

Keywords: Intelligent Autonomous Vehicles, Estimation, Sensors and Actuators
Abstract: This paper presents a low-cost, portable radar-based roadside system for vehicle tracking at intersections, near construction work zones, and on regular roads. The system automatically compensates for radar misalignment with the road and provides accurate per-lane tracking of vehicles in straight road scenarios. The proposed device integrates a millimeter-wave radar with an onboard Raspberry Pi controller and employs a modular processing pipeline combining Density-Based Spatial Clustering of Applications with Noise (DBSCAN) for clustering and an Unscented Kalman Filter (UKF) for multi-vehicle tracking. Experimental results on multi-lane straight road scenarios show that the proposed per-lane alignment achieves near-zero lateral mean squared error (MSE) and restores reliable and consistent lane assignment across the full sensing range, while an averaged transformation computed from the lane-specific corrections provides a simpler implementation with only modest loss of accuracy.

Keywords: Intelligent Autonomous Vehicles, Robotics, Machine Learning in modeling, estimation, and control
Abstract: Road traffic accidents are a leading cause of fatalities worldwide. In the US, human error causes 94% of crashes, resulting in excess of 7,000 pedestrian fatalities and 500 billion in costs annually. Autonomous Vehicles (AVs) with emergency collision avoidance systems that operate at the limits of vehicle dynamics at a high frequency, a dual constraint of nonlinear kinodynamic accuracy and computational efficiency, further enhance safety benefits during adverse weather, cybersecurity breaches and dangerous human driving evasion when AVs and human drivers share roads. This paper parameterizes a Deep Reinforcement Learning (DRL) collision avoidance policy Out-Of-Distribution (OOD) utilizing race car overtaking, without explicit geometric mimicry reference trajectory guidance, in simulation, with a physics-informed, simulator exploit-aware reward to encode nonlinear vehicle kinodynamics. Two policies are evaluated, a default uni-direction and a reversed heading variant which navigates in the opposite direction to other cars, that both outperform a Model Predictive Control and Artificial Potential Function (MPC-APF) baseline, with zero-shot transfer to proportionally scaled hardware, across three intersection collision scenarios, at 31× fewer Floating Point Operations (FLOPS) and 64× lower inference latency. The reversed heading policy outperforms the default racing overtaking policy in head-to-head collisions by 30% and the baseline by 50%, and matches the former in side collisions, where both DRL policies evade 10% greater than numerical optimal control.

Keywords: Machine Learning in modeling, estimation, and control, Human-Machine and Human-Robot Systems, Intelligent Autonomous Vehicles
Abstract: The increasing integration of artificial intelligence (AI) models into private and public sectors introduces an increasingly urgent concern for the accountability of machine models. This has caused a rapidly developing need for explainability, particularly for deep learning techniques such as convolutional neural networks (CNNs). One current explainable AI (XAI) method, Grad-CAM, struggles with interpretability and generalization of explainability across larger datasets. This study seeks to use Uniform Manifold Approximation and Projection (UMAP) to develop a methodology for explaining the decision process behind a CNN classification using low-dimensional embeddings, as well as compare the model’s “logic” to that of the human logic that determines the labels. This methodology was explored in relation to NASA air mobility integration objectives via examination of the classification of a vacant lot as suitable or unsuitable for placement of an air taxi stop. The dataset was developed based on Los Angeles' statutes and expert recommendations, since it was desirable to have embedded traceable logic in the data. The Convolutional Neural Network developed had an accuracy of 97.60%. UMAP clustering of Grad-CAMs based on the CNN yielded a vaguely clustered projection, the distribution of which suggested very little correlation between the human logic that created the dataset and the patterns developed in the UMAP space. This implies that the decision process of the model, as explained by the Grad-CAMs, differs significantly from the human decision process underlying the distributions of the datasets’ labels. The human-to-machine disconnect highlighted by this methodology shows a need for further development of a vision-based explanation analysis framework to ensure responsibility in their usage in machine learning solutions to problems across all industries.

Keywords: Nonlinear Control Systems, Intelligent Autonomous Vehicles, Control Design
Abstract: This paper introduces a new concept of resilient control Lyapunov function based quadratic program (RCLF-QP) framework that guarantees stabilization for nonlinear autonomous ground vehicles (AGVs) that may be subjected to a broad class of cyber attacks, including exponentially unbounded false data injection (EU-FDI) attacks on their control to input (C2A) channels. The proposed framework extends the standard robust CLF methodology and integrates a compensation term into the Lyapunov decrease condition with online real-time adaptation. This framework guarantees resilient stability, convergence, and operation without prior knowledge of the attack bounds, while the resulting controller enforces the decrease condition. A rigorous Lyapunov analysis establishes that the resulting closed-loop system guarantees resilient stability and convergence, with tracking errors remaining uniformly ultimately bounded (UUB) under wide range of time-varying and unbounded FDI attacks.

Keywords: Nonlinear Control Systems, Path Planning and Motion Control, Intelligent Autonomous Vehicles
Abstract: This paper studies the construction of control barrier functions (CBFs) for force-controlled nonholonomic mobile robots subject to relative-degree-two safety constraints arising from position-level obstacle avoidance. A scaling-based reciprocal barrier construction is proposed, in which a positive motion-dependent scaling factor is placed in the numerator of a reciprocal barrier associated with the original physical safety function. The resulting barrier is defined exactly on the interior of the physical safe set and becomes singular on its boundary, thereby preserving the certified interior domain of the original safety constraint while recovering first-order control authority. For a force-controlled nonholonomic robot model, sufficient conditions are derived under which the proposed construction defines a reciprocal CBF, and the interior of the physical safe set is forward invariant under controllers satisfying the induced reciprocal-CBF condition. The same construction mechanism is further illustrated for a scalar strict-feedback class under explicit structural assumptions. Numerical simulations demonstrate the induced safe-set geometry and its integration with an optimization-based control framework for obstacle avoidance.

Keywords: Path Planning and Motion Control, Unmanned Ground and Aerial Vehicles, Intelligent Autonomous Vehicles
Abstract: This paper introduces a new three-dimensional guidance law for a quadrotor using a dynamical systems based approach. The guidance method employs a spherical attractor class of dynamics, characterized by the desired radius, azimuthal angle, and zenith angle variations. The proposed guidance formulation achieves multiple objectives within a single framework, including convergence to a desired viewpoint on a spherical surface for point of interest inspection, spherical spiral trajectories for complete inspection, and target enclosure through continuous circumnavigation. A stability analysis establishes convergence of the quadrotor trajectories to the desired spherical surface from arbitrary initial positions. Numerical simulations validate the effectiveness of the proposed guidance method.

Keywords: Control Design, Nonlinear Control Systems, Control Applications
Abstract: This paper presents an online reinforcement-learning framework for safe gain scheduling of a nonlinear quadcopter controller. Rather than learning thrust and torque commands directly, the proposed method selects gain vectors online from a finite library of pre-certified stabilizing controllers, thereby preserving the structure of the underlying snap-based control law. Safety is enforced by restricting the policy to admissible gains that maintain forward invariance of a prescribed safe state set, while dwell-time constraints prevent excessively fast switching. To reduce the action-space dimension, translational gains are shared across spatial axes by exploiting the isotropic structure of the translational dynamics, whereas yaw gains are scheduled independently. A deep Q-network learns to adjust feedback authority according to the current flight condition, using aggressive gains during large transients and milder gains near hover. High-fidelity nonlinear simulations demonstrate accurate trajectory tracking, bounded attitude motion, reduced control effort near convergence, and stable hover regulation under online safe gain scheduling.

Keywords: Estimation, Modeling and Validation, Nonlinear Control Systems
Abstract: This paper presents an adaptive sliding-mode control (ASMC) strategy for robust regulation of spin dynamics in a double quantum dot (DQD) system subject to decoherence and hyperfine noise. The dynamics are modeled in the single-triplet subspace using a Bloch-sphere representation, where both the control input and hyperfine disturbance act through the same channel, resulting in matched uncertainties. A sliding manifold is designed to couple quantum coherence with population error, and an adaptive control law is developed to enforce finite-time convergence to this manifold. Lyapunov-based analysis establishes reachability, invariance, and boundedness of the disturbance estimate, ensuring robustness against time-varying hyperfine interactions. Numerical simulations confirm that the proposed ASMC framework achieves stable spin-state regulation and accurate population tracking despite model uncertainty. This work extends adaptive sliding-mode control to the regulation of double-quantum dot spin dynamics under hyperfine disturbance, offering a practical and resilient control approach for quantum dot spintronics that complements existing decoupling, error correction, and feedback control strategies.

Keywords: Estimation, Stochastic Systems, Nonlinear Control Systems
Abstract: A novel discrete-time reduced order extended Kalman filter was previously designed for stochastic systems as a locally optimal minimum variance estimator through quadratic minimization of the estimation error covariance. It is shown that this novel discrete-time reduced order extended Kalman filter has estimation error which is finite-time, mean-square bounded and has the H-infinity property to attenuate finite energy disturbances. Application of the filter to the nonlinear PMSM alpha-beta motor model is shown, with computational verification through simulation of the mean-square error boundedness and H-infinity property of the filter.

Keywords: Modeling and Validation, Motion and Vibration Control, Nonlinear Control Systems
Abstract: This paper introduces the thermomagnetic pendulum (TMP) as a coupled magneto-thermo-mechanical system, and develops a physics-based computational dynamics model. The TMP comprises of a permanent magnet and a ferromagnetic bob subjected to position-dependent heating. Thermal cycling near the Curie temperature weakens and restores magnetic attraction, producing cyclic forcing. The model couples transient three-dimensional heat transfer within the bob, a nonlinear temperature-dependent magnetization law incorporating demagnetization effects, and the mechanical dynamics of the pendulum. The distributed magnetic force is integrated over the bob to obtain a torque that depends jointly on the pendulum configuration and spatial temperature field. Finite element method-based simulations show rapid force reduction near the Curie point, angular asymmetry in the magnetic torque, and decreasing torque with increasing bob-magnet separation. Simulations with select initial conditions also indicate stable periodic motion of the TMP.

Keywords: Nonlinear Control Systems, Control Design
Abstract: The increasing demand for high-agility maneuvers in modern autonomous vehicles requires control frameworks capable of handling 6-DOF coupled translational and rotational motions. This paper addresses the problem of pose regulation and trajectory tracking PID control for a fully actuated rigid body vehicle directly on the tangent bundle TSE(3) associated with the special Euclidean group SE(3). The proposed approach avoids the singularities inherent in local parameterizations, including the use of exponential coordinates followed in previous strategies. Unlike traditional schemes that decouple translational and rotational motion, this framework respects the underlying Lie group structure, while integral control is incorporated using a dynamic state feedback approach. Lyapunov-based analysis guarantees almost global asymptotic stability, and the performance is validated through numerical simulations.

Keywords: Nonlinear Control Systems, Control Design
Abstract: Nonaffine nonlinear systems include a wide class of nonlinear systems, which impose great challenges for control design. Conventional approaches like the backstepping dynamic surface control techniques usually rely on dynamic inverse of the system in their design with inherent model dependencies. Thus, they cannot be implemented directly for nonaffine nonlinear system. In this paper, a tracking control method adopted based on the dynamic surface control is introduced. Specifically, it uses first-order filters and decouples the system in a form of quasi-independent cascade structure. The approach is fully model-free, uses full-state or partial-state information of the system, and can address matched and mismatched uncertainties and disturbances. Two numerical case studies are provided to examine the efficacy of the proposed method.

Keywords: Motion and Vibration Control, Control Design, Control Applications
Abstract: The sway oscillations of a variable-length pendulum is controlled through modulation of its cable length which induces a Coriolis damping force. Effective modulation requires precise synchronization with the pendulum motion, particularly at large sway amplitudes. In this study, it is demonstrated that the product of the measured sway angle and angular velocity facilitates such synchronization. A simple feedback control strategy is developed based on this observation. The proposed controller mitigates the need for large-amplitude length variations and prevents loss of string tension. The performance of the controller is validated through numerical simulations.

Keywords: Education for modeling, estimation and control, Modeling and Validation
Abstract: The key contribution of this paper is GoLab-C, a set of low-cost, portable control laboratory kits integrated with high-fidelity digital twins to support undergraduate and graduate control theory education. The platform addresses the disconnect between theoretical instruction and hands-on experimentation by enabling in-class demonstrations, laboratory activities, and self-paced learning through both physical systems and virtual simulations. The modular setups, including pendulum- and cart-based systems actuated by DC motors, allow students to design, implement, and analyze closed-loop controllers using real-time hardware and MATLAB Simscape and Simulink. Case studies demonstrate system modeling, parameter identification, and PID control design, highlighting practical challenges such as noise, nonlinearities, and implementation constraints. The combined use of physical experiments and digital twins provides multiple representations of learning, improving conceptual understanding and practical skills in control systems. Preliminary student assessment results indicate positive impacts on engagement, confidence, and learning outcomes.

Keywords: Modelling and Control of Environmental Systems, Chemical Process Control, Machine Learning in modeling, estimation, and control
Abstract: Predicting membrane fouling is challenging due to its dynamic and multifactorial nature, and existing approaches lack transparency or struggle to capture fouling trends across multiple operational cycles. In this work, transmembrane pressure, TMP, is used as the indicator of membrane fouling, and three modeling approaches are introduced, to enable accurate longterm predictions in an ultrafiltration process. The first uses symbolic regression for predicting ongoing TMP across filtration phases. The second applies random forest regression to predict and forecast initial TMP values. The third, a grey-box autoregressive model with dynamic parameter estimation, forecasts both ongoing TMP and initial TMP values.

Keywords: Biomechanical Systems, Assistive and Rehabilitation Robotics, Machine Learning in modeling, estimation, and control
Abstract: Human walkers encounter diverse terrains such as sand and solid ground, which result in distinct gait locomotion and energy costs. This study aims to design and validate a knee stiffness-based model predictive control approach for knee exoskeletons to assist walking on sand. A comparative analysis of human gait, kinematics, and kinetics on sand versus solid ground is conducted. A machine learning-based estimation scheme is developed to predict ground reaction forces in real time. These predictions, combined with human joint torque estimates, are used to design a knee exoskeleton controller that employs a model predictive stiffness control strategy. The experiments demonstrate significant differences in lower limb kinematics and kinetics, as well as a 15% reduction in major muscle activation and a 3.7% (albeit insignificant) reduction in metabolic cost during exoskeleton-assisted walking on sand. These results confirm that the proposed control framework effectively improves walking efficiency on sandy terrain by reducing both muscular effort and metabolic demand.

Keywords: Stochastic Systems, Machine Learning in modeling, estimation, and control, Adaptive and Learning Systems
Abstract: In this paper, we investigate the convergence dynamics of stochastic approximation algorithms under delayed gradient updates and their impact on asynchronous machine learning performance. We present a comprehensive case study comparing the asymptotic behavior of stochastic gradient descent with theoretical guarantees from stochastic approximation theory, examining both decaying and constant learning rate regimes. For decaying rates, we verify convergence but demonstrate that high delays necessitate impractical learning rates; for constant rates, we show that systematically exploiting the learning rate and delay relationship enables robust asynchronous learning. Furthermore, we identify that popular adaptive optimizers degrade under delays because of estimating curvature at current rather than delayed iterates. Finally, we propose a theoretically grounded Hessian-based correction framework using modified Newton-Raphson iterations that explicitly account for gradient staleness.

Keywords: Transportation Systems
Abstract: This paper investigates the difference flatness property of the discrete-time nonlinear METANET traffic model and leverages it for systematic freeway traffic control design. It is shown that traffic density and on-ramp queue constitute a suitable local flat output, from which traffic speed, ramp flow, and variable speed limit (VSL) inputs can be explicitly reconstructed using a finite number of forward shifts and measurable exogenous inputs. Exploiting this property, we develop and evaluate a flatness-based control strategy using real-world freeway traffic data, and benchmark it against the conventional ALINEA+VSL feedback controller. Simulation results demonstrate that the unconstrained flatness-based controller achieves nearperfect tracking of freeway density and speed, though it may occasionally produce infeasible ramp flows. Imposing operational constraints ensures feasible control actions while maintaining high tracking performance that still surpasses ALINEA+VSL. These findings highlight the practical benefits of difference flatness in freeway traffic management and provide a foundation for constraint-aware strategies, including model predictive control, for future research.

Keywords: Modeling and Validation, Discrete Event Dynamic Systems, Control Applications
Abstract: This paper studies secure simultaneous wireless information and power transfer (SWIPT) in vehicle-to-infrastructure (V2I) systems from a system-level perspective. Building on classical interference-coupled power-control theory, the secure V2I SWIPT model is formulated in a power-flow form that incorporates multi-user interference, receiver noise, artificial-noise leakage, and energy-signal leakage. The resulting power recursion is interpreted as an interconnected discrete-time system, and the standard spectral-radius condition is applied to characterize finite-power feasibility and convergence. The formulation is incorporated into a secure resource-allocation problem with legitimate-user signal-to-interference-plus-noise ratio (SINR), harvested-energy, and secrecy constraints, and is further extended to multi-user multipleinput multiple-output (MU-MIMO) systems. Numerical results illustrate the effects of SINR requirements, interference coupling, artificial noise, energy delivery, and secrecy constraints on system feasibility and transmit power.

Keywords: Mechatronic Systems, Modeling and Validation, Robotics
Abstract: This work presents an experimentally grounded modeling and gait-optimization framework for a compliant worm robot traversing corrugated pipes. A hybrid locomotion model captures continuous motion within a groove and discrete anchoring transitions, while a slack-aware actuation model maps the commanded gait to the realized body-length change. A calibrated energy model predicts source-side power consumption. These models are integrated into a multi-objective optimization that maximizes average speed while minimizing average power and incorporates a kinematic margin to improve the hardware deployability of optimized gaits. Experiments validate the component models and the sim-to-real transfer of representative robust gaits.

Keywords: Mining, Mineral and Metal Processing, Machine Learning in modeling, estimation, and control, Modeling and Validation
Abstract: This work describes, analyzes, and models a characteristic, repeated structure in the model-plant mismatch in hot-dip galvannealing of steel strips. A potential cause of the residuals of existing models is proposed using a combination of known metallurgical phenomena and observations from industrial data. Three data-driven models are shown to accurately predict the residuals, with the aim of improving model-based coating weight control.

Keywords: Modeling and Validation, Chemical Process Control
Abstract: A non-linear, dynamic model is developed for a heat exchanger network used in a cryogenic separation process unit which separates synthesis gas into various product and intermediate streams. The model aims to capture non-linearities presented by heat transfer rates, and flashing with limited plant measurements. This model is validated against industrial plant data. The heat exchanger network is presented as separate subsystems, but can be integrated for future work. The proposed dynamic, non-linear model of the system will aid in the design of control strategies for the control and optimisation of the process unit.

Keywords: Estimation, Mechatronic Systems, Modeling and Validation
Abstract: State estimation becomes particularly challenging for systems with linear process dynamics but highly nonlinear, non-monotonic, and hysteretic measurement maps. This paper proposes a unified estimation framework that combines Preisach-based hysteresis compensation with a region-wise nonlinear observer to enable accurate and provably stable state reconstruction under these conditions. The approach is demonstrated on magnetic position estimation for a long-stroke ferrous hydraulic actuator, where a permanent magnet attached to the piston is measured by distributed anisotropic magnetoresistive sensors along a 660 mm stroke. A discrete Preisach model with constrained least-squares identification captures major and minor hysteresis loops in each measurement channel, while a smooth polynomial baseline isolates the underlying hysteresis-free field–position relationship. The compensated measurements are processed using a switched-gain observer whose gains are synthesized via linear matrix inequalities over slope-bounded regions of the non-monotonic output map. The LMI-based gain design provides exponential convergence within each region, while the hysteresis-aware switching logic reduces chattering near region boundaries. Experimental results under randomized motion profiles and large initial errors demonstrate root-mean-square position errors below 2 mm and maximum errors below 1% of the full stroke, significantly outperforming observers based on averaged magnetic field models. The resulting estimator requires only a small number of gain matrices and low-order hysteresis models, making it suitable for embedded implementation.

Keywords: Modeling and Validation, Motion and Vibration Control, Mechatronic Systems
Abstract: Despite decades of research, assistance functions for tower cranes have not yet become state of the art in industry, mainly due to the lack of testing on industrial products. Due to their lightweight structure, tower cranes are susceptible not only to payload oscillations but also structural oscillations. When hydraulic drives are used, structural excitation is propagated back to the hydraulic system and oscillations are amplified. To address these challenges, model-based control requires adequate modeling of the coupled dynamics. The majority of existing research uses simulations or lab-scale cranes. In addition, hydraulic dynamics are often linearized for control design. This paper presents a dynamic model of a luffing crane that combines a flexible multibody mechanical system with a comprehensive dynamic model of the hydraulic luffing mechanism. The model considers the nonlinear dynamics of the load control valves, including variable dead times. By incorporating a Prandtl-Ishlinskii model, the hysteresis of the main valve is considered and can be directly inverted for control design. Additionally, the influence of the oil temperature on the flow velocity is considered with a model-based approach. Using measurement data from an industrial crane, unknown parameters are identified and it is demonstrated that the coupled dynamics can be represented adequately. The resulting comprehensive modeling approach, considering varying dead times caused by the load control valves, as well as hysteresis and oil temperature effects in the main valve, can be transferred to comparable applications and used as a simulation framework to validate control designs.

Keywords: Motion and Vibration Control, Modeling and Validation, Marine Systems
Abstract: Large-scale deployment of floating offshore platforms requires dynamic power cables that withstand site-specific offshore loading. These cables are vulnerable to vortex-induced vibrations (VIV), which increase cyclic loading and accelerate fatigue damage. This study investigates High Modulus Polyethylene (HMPE) sleeves as a stiffness-enhancement strategy for critical cable locations. MoorDyn simulations of two sleeve configurations show that sleeve placement influences fatigue distribution: a central sleeve reduces touch-down fatigue damage, while sleeves at both cable ends reduce hang-off fatigue damage but increase touch-down damage. These results demonstrate the importance of sleeve placement for fatigue mitigation.

Keywords: Modelling, Identification and Signal Processing, Motion and Vibration Control, Sensors and Actuators
Abstract: Bearing fault diagnosis at low rotational speeds is difficult when strong mechanical noise masks fault-related impulses. This study investigates robust bearing-fault detection under such severe conditions, and compares the diagnostic performance of a piezoelectric jerk sensor with that of a conventional acceleration sensor when both are combined with two-stage variational mode decomposition (VMD). The dominant interference considered here is the noise generated by a chain transmission system, where the vibration amplitude is comparable to or higher than the impulses produced by small faults. The vibration signals were measured at the housing of the bearings with localized faults on the inner race. The proposed processing applies VMD sequentially: the first stage isolates a candidate intrinsic mode function (IMF) that contains the fault-sensitive high-frequency component, and the second stage re-decomposes that IMF with tighter bandwidth control to suppress residual mechanical noise and isolate weak fault-induced impulsiveness. Across the tested cases, the fault-related pulsatile response exhibited a dominant component at approximately 55 kHz in the case of an inner-race fault. When the second-stage VMD targets the first-stage IMF with a center frequency closest to this reference component, the two-stage approach reliably isolates weak fault waveforms, provided that the parameters of the VMD are properly selected. The comparative results show that the jerk sensor, when integrated with the proposed two-stage VMD, yields clearer fault impulsiveness and more distinct fault periodicity than the acceleration sensor for inner-race faults.

Keywords: Motion and Vibration Control, Control Design, Control Applications
Abstract: This study investigates active vibration control that exploits seismic waveforms observed at remote sites. We propose a 2-DOF control framework that directly incorporates waveform data from multiple remote locations. The system integrates an LQR based feedback controller with a feedforward controller generated from locally and remotely observed waveforms processed through optimally designed signal processing filters. A sample-by-sample switching strategy selects the most effective feedforward input. Design parameters, including filter characteristics and switching rules, are optimized via particle swarm optimization (PSO) with simulated responses to several major earthquakes. Simulation results demonstrate the effectiveness of the proposed approach.

Keywords: Electromechanical systems, Modeling and Validation, Motion and Vibration Control
Abstract: Temporally modulated piezoelectric metasurfaces can produce multi-band flexural wave attenuation from spatially uniform resonator arrays, yet predictive modeling of such systems requires frameworks that resolve distributed electromechanical coupling and finite-structure boundary effects. This paper presents a three-dimensional finite element model of a piezoelectric metasurface beam with time-varying synthetic inductance, developed in COMSOL Multiphysics 6.2. Five shunted piezoelectric patches on an aluminum beam are driven by a sinusoidally modulated inductance at fm = 1100 Hz, and an open-circuit reference simulation removes frequency-dependent probe artifacts from the numerical transmission ratio. A parametric study across different inductance modulation ratios between 0.20 and 0.35 reveals the progressive redistribution of attenuation energy from the primary bandgap to Floquet sidebands with increasing modulation amplitude. The crossover between center-dominated and sideband-dominated regimes occurs near Lm/L0 = 0.25. At higher ratios, second-order sidebands emerge and the effective center frequency drifts upward due to the nonlinear time-averaging of L(t). The model also captures a dispersive asymmetry between upper and lower sidebands inherent to Euler–Bernoulli beam dynamics. These findings establish the three-dimensional FEM approach as a predictive tool for designing temporally modulated adaptive metastructures.

Keywords: Motion and Vibration Control, Modeling and Validation, Modelling, Identification and Signal Processing
Abstract: A tensegrity structure consists of compression elements called bars or struts that do not contact each other, combined with tensioned cables. This combination of tension and compression elements contributes to the system's stability by forming a self-balancing force network. Tensegrity structures have high strength-to-weight ratios and foldability, enabling wide applications across engineering fields, including aerospace, civil, and mechanical engineering. Due to the widespread use of tensegrity structures, a variety of studies have examined the dynamic characteristics of tensegrity systems; however, many have proven inadequate for various reasons. To reduce computational cost and account for bending effects, this manuscript developed a framework based on the high-order finite element method. The framework models the cables as truss elements with a quadratic shape function and the bars as frame elements based on the Euler-Bernoulli beam theory, and it uses partitioning to release the moment at the end of the elements. The method has demonstrated very good accuracy, significant reductions in computational cost, and the ability to perform bending analysis of structural bar members.

Keywords: Motion and Vibration Control, Modeling and Validation, Transportation Systems
Abstract: This study aims to enhance and automate the evaluation of railway operational safety by developing a machine learning model that estimates rail irregularities from PQ measurement data. The model was trained using a diverse set of synthetic training data generated through simulations, and its effectiveness was evaluated using both simulation data and real-world measurement data. The results indicate that the proposed approach achieves satisfactory performance under simulation-based conditions. However, when validated against real measurement data, several limitations were observed. These discrepancies are considered to stem from the gap between simulated training data and real-world phenomena. Therefore, further improvements are required, including the refinement of training data set and the incorporation of additional informative features.

Keywords: Human-Machine and Human-Robot Systems, Robotics
Abstract: Admittance control is widely used for human-robot collaborative transport tasks, where the robot senses and responds to external forces applied by a human. For co-transport of rigid objects, force-based admittance control is a standard paradigm: the robot regulates its motion to emulate a desired mass-damper behavior in response to forces measured at the end effector. This enables compliant, intuitive interaction and has been demonstrated in a number of co-transport settings. An advantage of admittance controllers is that they use typically-available end-effector force sensors and do not require additional vision-based sensors. However, the challenge when the manipulated object is flexible is that the object dynamics distorts the applied force sensed by the robot and thereby limits the performance of admittance controllers. The main contribution of this work is to improve the performance of current admittance controllers during flexible-object co-transport by (i) estimating the human action using an object-dynamics model and then (ii) using the estimated human action to iteratively correct the admittance-controller error. The proposed approach is validated on a UR10e industrial robot platform with seven human subjects, and experimental results show a 79% reduction in root-mean-square (RMS) deformation versus admittance control alone

Keywords: Machine Learning in modeling, estimation, and control, Modeling and Validation, Estimation
Abstract: This paper studies parameter identification and trajectory prediction of nanowires suspended in fluid under externally applied electric fields. A physics-informed neural network (PINN) framework is proposed to integrate electrophoretic dynamics with data-driven learning, enabling simultaneous estimation of unknown zeta potentials and prediction of nanowire motion. The model incorporates governing equations into the loss function and takes time and electrode control inputs as arguments. For parameter identification, the PINN achieves high accuracy and outperforms a least-squares baseline on both simulation and experimental data. For trajectory prediction, the method is evaluated on experimental datasets with both fixed and time-varying control inputs. The PINN consistently achieves lower prediction error and variance than a feedforward neural network and a physics-based electrophoretic model. Notably, it generalizes well to unseen trajectories and remains robust in challenging regions with nonlinear electric fields. These results demonstrate that the proposed framework effectively addresses model mismatch and parameter uncertainty, providing a unified and robust approach for nanowire modeling and prediction.

Keywords: Path Planning and Motion Control, Intelligent Autonomous Vehicles, Robotics
Abstract: Sampling-based motion planners are widely used for autonomous driving in complex environments because of their scalability and probabilistic completeness. However, the resulting paths are typically piecewise linear, exhibiting discontinuities in curvature and poor dynamic feasibility, which limits their applicability for high-performance motion execution. This paper introduces Elastic Curve Smoothing (ECS), a physics-informed path smoothing method that models trajectories as deformable elastica curves. By formulating smoothing as a dynamic system, ECS enforces curvature continuity, obstacle avoidance, and non-holonomic constraints through physically grounded forces. In contrast to existing methods, ECS provides a unified mechanism to generate globally smooth paths with bounded curvature, without relying on heuristic tuning or local approximations. The resulting path is further utilized for time-optimal trajectory generation, demonstrating improved optimization convergence and reduced traversal time. Simulation results demonstrate superior smoothness, curvature consistency, and traversal efficiency compared to classical methods, while maintaining real-time feasibility. Experimental results validate the effectiveness of the proposed pipeline, demonstrating accurate tracking performance in challenging scenarios with tight turns and obstacle constraints.

Keywords: Robotics, Mechatronic Systems
Abstract: Dexterous hands with humanoid fingers and a high number of degrees of freedom (DOFs) are desirable for robots to effectively serve at homes and workplaces. Despite their popularity, tendon-driven designs suffer from poor reliability and durability, as well as lack of bi-directional actuation. Linkage-driven designs can potentially avoid such limitations. However, they face challenges on integrating multiple DOFs to each finger without using complicated mechanisms that require a great number of miniature moving parts. This paper presents a new design of linkage-driven fingers. Thanks to a unique design of the MCP joint, with a lot fewer moving parts than existing designs, the finger realizes three bi-directionally actuated DOFs that are easily backdrivable. In addition, actuation of the 2-DOF MCP joint and the 1-DOF coupled PIP-DIP joints is completely decoupled. These advantages greatly simplify the kinematics and, in turn, benefit sensing and control. The proposed design is optimized through kinematics simulation. A prototype hand of three fingers is fabricated and physically tested for finger strength, grip strength, grasp dexterity, and tool usage.

Keywords: Robotics, Mechatronic Systems, Sensors and Actuators
Abstract: Handheld demonstration interfaces are becoming an increasingly popular choice to collect data for robot learning manipulation skills from human demonstrations. They mechanically resemble the actual robot end-effectors and are handheld and actuated by mentors to perform demonstration actions. Despite their advantages over conventional paradigms such as motion capture of native human actions and teleoperated demonstrations, such interfaces face unanswered design problems. In particular, tactile feedback to the users is missing from existing designs. Dexterity is usually achieved with complex designs that suffer from various weaknesses. This paper presents a new design that provides a better balance between physical simplicity and dexterity, together with a novel method to provide tactile feedback to the users. A fingernail structure with a flexion degree of freedom (DOF) is added to conventional single-DOF grippers, so as to enable fine motor skills that require pincer grasp. The addition allows a user of the interface to handle a variety of commonly encountered manipulation tasks in our lives while keeping the interface strategically simple. We also propose a novel acoustic component to convert subtle contact events to audio feedback that can be effectively interpreted by the users. The component can be made rugged at a low cost, offering an efficient tactile feedback solution for handheld demonstration interfaces. A user study is conducted to examine the effectiveness of the two new features on dexterity and tactile perception.

Keywords: Soft Robotics, Modeling and Validation, Estimation
Abstract: Pneumatic soft robotic actuators exhibit highly nonlinear, pressure-dependent, and hysteretic behavior, making accurate dynamic modeling difficult with purely analytical approaches. Although neural networks can learn these dynamics directly from data, purely data-driven models often require longer training and provide limited physical interpretability. This paper presents a physics-informed recurrent neural network (PIRNN) for modeling a pneumatic fabric-based soft robotic arm. The proposed model combines a linear parameter-varying (LPV) physics model with a gated recurrent unit (GRU) network that learns residual dynamics not captured by the physics model. The model predicts the actuator bending angle and arc length from measured chamber pressure. Experimental results on unseen test trajectories show that the PIRNN achieves short-term prediction accuracy comparable to that of a purely data-driven recurrent baseline, reducing the epoch requirement by 45.6% and the total training time by approximately 28%. In addition, the proposed model demonstrates improved long-term prediction stability, maintaining bounded behavior over extended rollouts compared to the purely data-driven baseline. These results indicate that embedding physics models within a recurrent learning framework can improve training efficiency, maintain predictive accuracy, and enhance long-term prediction stability by reducing error accumulation in extended rollouts.

Keywords: Automotive Systems, Optimal Control, Modeling and Validation
Abstract: Rule- or logic-based energy management strategies (EMSs) remain the dominant approach in production hybrid electric vehicles (HEVs) for their simplicity and compatibility with existing control architectures. Control rules for such strategies are commonly extracted from optimization-based algorithms such as dynamic programming (DP). However, most DP formulations optimize fuel economy over a single driving cycle without accounting for phase-dependent behavior that governs the U.S. EPA 5-mode certified fuel economy (FE) rating. Furthermore, standard DP formulations do not explicitly enforce the net energy change fraction ratio (NECFR) criterion for a valid charge-sustaining test. This paper proposes a nested optimization framework where phase-specific cost weights and terminal state-of-charge are determined by an outer loop, whereas DP solves the inner energy management problem. In this way, the proposed method explicitly searches for phase-wise net energy change (NEC) trajectories that maximize the 5-mode FE while complying with the NECFR criterion. For the considered transmission-mounted electric device (TMED) HEV, the proposed framework produces NECFR-compliant DP solutions which can be used as benchmark for EMS development. The proposed method yields modest improvements in 5-mode combined FE by 0.42%, 0.21%, and 0.33% in the city, highway, and combined FE, respectively, compared with a typical DP solution with phase-end SOC constraints.

Keywords: Transportation Systems, Automotive Systems, Intelligent Autonomous Vehicles
Abstract: This paper develops a control system for fuel cell electric bus (FCEB) platooning to ensure safe operation and improve energy conversion efficiency under realistic intercity driving conditions. The proposed hierarchical control framework consists of an upper-level controller for safe spacing regulation and smooth speed coordination, and a lower-level controller for optimal power allocation between the fuel cell system and the battery. A two-bus platoon with identical powertrain configurations is evaluated using an intercity highway driving cycle derived from Alberta Highway 2 between Edmonton and Leduc. Simulation results show that the designed control system enables safe platooning operation and reduces energy consumption. The follower FCEB achieves energy savings of up to 6.7% at a regulation-compliant time gap of 2 sec. These results demonstrate the potential of coordinated platoon control and powertrain energy management for improving the energy efficiency of intercity FCEB operations.

Keywords: Cyber physical systems, Estimation, Power and Energy Systems
Abstract: Modern power grids are increasingly vulnerable to coordinated false data injection attacks (FDIAs) designed to evade conventional residual-based detection. In particular, dynamic attacks can remain stealthy at individual sampling instants while introducing structured temporal deviations that adversely affect state estimation. This presentation summarizes the contributions and results of our recent paper, “Functional Subspace Projection for Detection of Coordinated Stealthy Attacks in Power Systems,” accepted for publication in the ASME Letters in Dynamic Systems and Control (Omiloli et. al 2026).

The work develops a joint framework for modeling and detecting coordinated dynamic stealthy attacks in multi-area power systems. A time-aggregated attack model captures the temporal evolution and inter-area coordination of attack signals while incorporating residual-stealthiness and sparsity constraints. For detection, a kernel-embedded functional subspace detection (KEFSD) method models residual trajectories in a reproducing kernel Hilbert space (RKHS) and employs RKHS-constrained functional principal component analysis to learn a nominal functional subspace. Online residual trajectories are projected onto this subspace, enabling detection of structured temporal deviations that may remain undetected by conventional instantaneous residual-based methods.

The framework is evaluated on a modified IEEE 14-bus system partitioned into three areas under coordinated stealthy attacks. At matched false-alarm conditions, KEFSD achieves area-under-the-curve (AUC) values of 84.8%, 83.0%, and 82.4% across the three areas, compared with 74.5%, 77.4%, and 75.8% for a conventional 2-norm detector and 55.0%, 53.0%, and 53.0% for a principal component analysis (PCA)-based detector. Sensitivity analyses further evaluate the effects of attack frequency, duty cycle, measurement noise, sliding-window size, and RKHS subspace dimension. The results demonstrate that incorporating temporal and functional structure into residual analysis improves the detection of coordinated stealthy attacks and provides robust performance across a range of operating and attack conditions.

Keywords: Power and Energy Systems, Adaptive and Learning Systems, Control Design
Abstract: We propose a novel extremum seeking control (ESC) method that operates in a lifted Koopman state space to minimize the filtered RMS energy in the dominant subspace. The lifted representation provides linear embeddings of nonlinear dynamics, enabling more accurate gradient estimation and dampening of state interference for more consistent ESC performance. Applied to a parameterized, forced, and time-varying non-autonomous Van der Pol oscillator, we show that the approach yields faster and more robust performance than operating ESC on the measured states. These advantages position the method for a diverse range of applications such as subsynchronous oscillation mitigation in inverter-dominated power systems.

Keywords: Power and Energy Systems, Control of Smart Buildings and Microgrids, Multi-agent and Networked Systems
Abstract: This paper presents a capacity-aware distributed control framework for an islanded AC microgrid with inverter-interfaced distributed generators (DGs) of unequal storage capacities. The scheme combines droop-based primary control, distributed secondary voltage restoration, and an SoC-aware droop adaptation layer. The measured load RMS current is used to generate SoC-based current-sharing references, which are mapped into active-power droop coefficients so that higher-capacity DGs supply more load while lower-capacity units are relieved. A unified model capturing fast electrical dynamics and slow SoC evolution is used to analyze the closed-loop behavior over two time scales. Simulation results demonstrate rapid voltage restoration, accurate power balance, capacity-consistent current and active-power sharing, and convergence of the SoC trajectories to a common profile. The results confirm effective coordination of fast-scale voltage regulation and slow-scale energy management in storage-rich islanded microgrids.

Keywords: Power and Energy Systems, Security and Privacy, Cyber physical systems
Abstract: This paper proposes a region-based approximation framework for detecting false data injection attacks in cyber-physical power systems. Drawing on domain decomposition principles, the framework partitions the operating space into regions and trains dedicated experts for each subproblem, reducing heterogeneity and improving detection sensitivity. Evaluated on the IEEE 300-bus system using real-world load data, the method improves the F1-score from 39.61% in traditional deep-learning results with a deep network to 78.39%, while maintaining operationally low false-alarm rates. Experiments with multiple routing strategies confirm the scalability and generality of the approach.

Keywords: Healthcare systems, Estimation, Biomechanical Systems
Abstract: This paper focuses on the estimation of ground reaction forces (GRFs) during walking using a single trunk-mounted inertial measurement unit (IMU) sensor. First, the real-time forward and side bending tilt angles are estimated for the IMU using a switched-gain nonlinear observer. The nonlinear observer is computationally efficient and uses off-line computation of observer gain while providing stability within a broad observer operating domain. The tilt angles are used to compute the anterior-posterior and vertical (gravity-aligned) acceleration components. The relationship between trunk acceleration and GRF is assumed to be a linear dynamic model and is represented using a discrete-time non-recursive equation with unknown filter coefficients. The filter coefficients are identified for each subject using training data. A noncausal filter identification using Wiener least squares estimation with regularization and constraints is utilized for training. The developed algorithm is then evaluated using our laboratory data and validated on the public EPIC biomechanics dataset. Analysis of the experimental data shows that, across EPIC subjects, the subject-level normalized rootmean-square error ranged from 2.7% to 4.2% for the anterior-posterior GRF component and from 2.9% to 6.1% for the vertical component. The results demonstrate that subject-specific estimation of total AP and V GRFs during steady treadmill walking can be achieved using a single low-cost wearable IMU sensor.

Keywords: Healthcare systems, Machine Learning in modeling, estimation, and control, Human-Machine and Human-Robot Systems
Abstract: Stroke survivors with hemiparesis often adopt compensatory movements that reinforce non-physiological patterns and hinder functional recovery. Telerehabilitation enables therapy outside the clinic without direct supervision, but requires automated detection of such compensatory movements. This study develops a machine learning approach to detect compensatory movements from data recorded by a Kinect sensor. Movement data from seven stroke patients were heuristically labeled using trunk and spinal motion features. A coarse decision tree classifier was trained on 30 kinematic features, achieving 92.3% accuracy in training and 93.8% accuracy on out-of-sample data. These preliminary results demonstrate the feasibility of detecting compensatory movements, representing a first step toward automated, real-time feedback in telerehabilitation applications.

Keywords: Healthcare systems, Modelling and Control of Biomedical Systems, Control Applications
Abstract: Prior developments of closed-loop controlled critical care treatments focused on the administration of individual treatments. However, how multiple closed-loop controlled critical care treatments interact with each other when used simultaneously has not been rigorously elucidated. Emerging efforts tend to use multiple individually developed closed-loop controlled critical care treatments without considering the interactions. We examined the characteristics of multiple uncoordinated closed-loop controlled critical care treatments using fluid resuscitation-intravenous anesthesia as a case study. Based on the steady-state analysis, we demonstrated that the set of reachable treatment targets (defined as the treatment targets satisfying prescribed constraints in the steady-state) for a given initial patient state is limited by pre-defined constraints that can be set to support patient safety. Based on the dynamic simulation analysis, we demonstrated that the patient response trajectory from an initial patient state to a reachable treatment target can be subject to transient violations of pre-defined constraints depending on the relative aggressiveness of individual closed-loop controlled critical care treatments. In sum, our analyses show an approach to jointly study individually developed closed-loop controlled critical care treatments that are simultaneously applied.

Keywords: Modelling and Control of Biomedical Systems, Biomechanical Systems, Healthcare systems
Abstract: Upper-extremity exoskeletons aim to assist human mobility but often fail to accurately represent human muscle behavior. To address this, we developed a model predictive control (MPC) framework that accounts for human effort through muscle co-contraction in a simulated muscle-driven system. The controller was applied to a one degree-of-freedom arm linkage to maintain setpoint tracking while reducing individual muscle forces. A constrained optimization cost function penalized tracking error and muscle length. Results demonstrate that including a muscle length penalty reduced tendon force for both muscles while maintaining sufficient tracking accuracy, demonstrating improved representation of human effort in assistive device design.

Keywords: Robotics, Healthcare systems, Modeling and Validation
Abstract: Vitreoretinal surgery demands sub-millimeter precision that exceeds human capabilities. While robotic platforms like the Steady-Hand Eye Robot (SHER) 3.0 enhance surgical dexterity and suppress physiological hand tremor, clinician training on these physical systems remains severely constrained by high cost, limited availability, and hardware fragility. To address this critical training bottleneck, we present a kinematically accurate virtual training environment. Developed using the Asynchronous Multi-Body Framework (AMBF), our simulator models the robot’s complex parallel kinematics, including over-constrained mechanisms, real-time rigid-body physics, closed-loop teleoperation control, and integration of an input device. A pilot user study involving teleoperated micro-reach tasks demonstrates that the proposed simulation environment provides a realistic and intuitive kinematic experience that is functionally comparable to the physical robot, thereby establishing a robust foundational training platform for advanced microsurgical procedures.

Keywords: Robotics, Path Planning and Motion Control, Healthcare systems
Abstract: Underactuated upper limb exoskeletons present a fundamental trajectory planning challenge: with fewer actuators than degrees of freedom in the human arm, conventional point-wise inverse kinematics (IK) methods produce discontinuous joint motions that are unsuitable for rehabilitation. This paper introduces Trajectory Optimization via L-BFGS (TOL), a framework that optimizes the entire joint trajectory simultaneously rather than solving each waypoint independently. By formulating a unied cost function over tracking accuracy, motion smoothness, and joint feasibility, and computing exact gradients through automatic dierentiation via the array-oriented numerical computation library JAX, TOL avoids the numerical instabilities and motion discontinuities inherent in Jacobian-based approaches. A warm-start initialization strategy further ensures reliable convergence within the constrained configuration space of the robot. In a pilot study with a 4-DOF underactuated rehabilitation exoskeleton and four healthy subjects across geometric and activities of daily living (ADL) tasks, TOL achieves a 4.6× improvement in end-effector tracking accuracy and 3370% reduction in motion jerk relative to Jacobian Inverse Kinematics (J-IK) and Damped Least-Squares (DLS) baselines, demonstrating its potential for generating clinically viable rehabilitation trajectories.

Keywords: Machine Learning in modeling, estimation, and control, Automotive Systems, Modeling and Validation
Abstract: In this paper, an adaptive ensemble-based method is proposed for dictionary selection in Koopman modeling of vehicle heating, ventilation, and air conditioning (HVAC) systems. Koopman-based models provide a computationally efficient framework for approximating nonlinear system dynamics, but their performance depends strongly on the choice of dictionary functions, which are often selected heuristically and tend to be oversized. The proposed approach combines bootstrap-based ensemble generation with hyperparameter optimization, allowing dictionary inclusion to be treated as part of the model selection process.

The method is evaluated on a Koopman model of an HVAC condenser subsystem using data generated from a high-fidelity GT-Suite simulation. Results show that the proposed approach reduces the dictionary size from seven to five basis functions while maintaining comparable predictive accuracy relative to a baseline full-dictionary Koopman model. Additional experiments with noisy training data demonstrate consistent selection of the same reduced dictionary across repeated trials, indicating robustness to measurement noise.

Keywords: Machine Learning in modeling, estimation, and control, Modelling, Identification and Signal Processing, Estimation
Abstract: The safe and reliable management of lithium-ion batteries (LiBs) requires accurate models that captures their complex dynamics and adapt to ageing and use conditions. Hybrid modeling tech niques that combine the explainability of reduced-order physics-based models with the expressivity of deep neural networks are emerging as viable solutions for capturing this complexity while directly using available data streams from the LiBs. However, there are still computational challenges with updating both components of a hybrid model with on board compute resources typically available on vehicles and systems that use the LiBs as energy source. In this paper, we propose an asynchronous learning scheme for updating the physics-based and data-driven components of the hybrid model. Specifically, we employ online system identification algorithms for real time updates of the prioritized physics-based component, and update the neural data-driven component with long horizon data as opportunities arise. The hybrid model employing this scheme is evaluated using laboratory experimentation with three common LiB cell chemistries. The results show that the approach offers robust model learning performance across the tested LiB chemistries at different ambient temperature conditions.

Keywords: Power and Energy Systems, Automotive Systems, Modeling and Validation
Abstract: Physics-based battery models (PBMs) are crucial for fast charging and degradation control, but most high-fidelity PBMs are too complex for real-time applications. Single-particle models (SPMs) are efficient, but they assume uniform behavior across electrode thickness, reducing accuracy at high C-rates significantly. This work proposes a spatially-resolved SPM that captures variations in solid stoichiometry, reaction current, and overpotential across porous electrodes while retaining low complexity. The model combines a conventional SPMe with spatial corrections from perturbation analysis and linear time-variant filters. This formulation adds only two additional states per electrode, yet estimates key electrochemical gradients and is provably stable. Simulation results show that our "SPMe+" reduces voltage prediction error by 65 to 90% (depending on the current profile) compared with a conventional SPMe, adding only 3 to 11% to CPU time in an embedded deployment, making it a valuable contribution to next-generation battery controls.

Keywords: Power and Energy Systems, Control Applications, Modeling and Validation
Abstract: This paper explores the coupling of dissimilar dynamic models for system-level analysis. Design, control, and estimation for dynamic systems require accurate and analytically tractable representations. However, development of such representations is impeded by the heterogeneous modeling paradigms used for subsystems and components. Prior efforts to address this provide frameworks to link model forms together, though these can neglect integration of data-driven models or relinquish features that support analysis. This work addresses this by introducing a control-oriented framework to couple two types of representations: physics-based models, and a class of data-driven models. The framework consists of a multistep procedure that transforms subsystem models into matching representations, constructs a fully coupled model using defined coupling terms, and analyzes control properties. A physics-based microgrid with a data-driven data center load model is used to demonstrate the proposed methodology and the foundation it provides to preserve analytical tractability. Results from the case study show that the coupling structure influences the equilibrium points and stability properties of the integrated system.

Keywords: Power and Energy Systems, Electromechanical systems
Abstract: Mechanical effects play an important role in solid-state batteries, where lithiation-induced particle swelling can substantially alter local transport and stress fields. Capturing these effects using a physics-based integrated formulation enables a unified description of the coupled interactions between the pressure applied at the electrode level and particle-scale lithiation and delithiation. The proposed approach combines spherical diffusion in the active particles, a small-strain, spherically symmetric elastic description of particle deformation and stress, and the representative volume element formulation to synthesize an overall electrode strain. However, particle volume changes within the Fickian diffusion equation lead to a moving-boundary problem and to poorly conditioned algebraic equations. To address this computational issue, this paper introduces a numerical-oriented approximation to preserve tractability, accounting for particle volume change in the solid-mechanics formulation while retaining a constant particle radius in the molar flux term of Fick’s law. A 2C discharge was simulated in PyBaMM, incorporating the integrated lithiation-induced stress model. The results show peak hydrostatic stresses of approximately 200 MPa in the positive and 500 MPa in the negative electrode, with corresponding surface displacements of 75 nm and 177 nm, respectively. Analyzing the effects of external pressure and C-rate on the electrodes reveals that the cathode response remains nearly insensitive to the investigated pressure range, while being more strongly influenced by C-rate. In contrast, the anode is significantly affected by externally applied pressure, except for low stack pressures, where the effect of C-rate is predominant. Overall, the results are consistent with the literature and demonstrate a practical compromise between physical fidelity and numerical robustness, making the model suitable for large-scale simulation and design optimization of solid-state batteries.

Keywords: Adaptive and Learning Systems, Power and Energy Systems, Control Design
Abstract: This paper proposes a control architecture integrating adaptation with Lyapunovbased Reference Governors (LRGs) to ensure state constraint satisfaction for first-order systems with parametric uncertainties. Adaptation combined with LRGs guarantees stability, ensures good control performance, and remains safe even with parametric uncertainties. Simulations of the fuel cell temperature regulation problem demonstrate that the proposed control architecture successfully meets all control and safety objectives, whereas the standard adaptation fails to achieve the latter.

Keywords: Linear Control Systems, Electromechanical systems, Sensors and Actuators
Abstract: The Power Amplifier Direct Current (PADC) dynamics uncertainty is evaluated in frequency response to study the impact on the servo loop of Short Stroke stage used to position lithography scanning system. The variation could be due to manufacturability or control attributes. The variation was investigated in frequency domain of both Magnitude and Phase. The open-loop and sensitivity functions of stable servo loop were examined under PADC uncertainty along with time domain evaluation under aggressive setpoint. The simulation results were followed by experiments suggested that the large variation of PADC phase can translate to minor variation on servo frequency response but yields significant delay impacts settling of the Stage under aggressive setpoint.

Keywords: Linear Control Systems, Modelling, Identification and Signal Processing, Mechatronic Systems
Abstract: In this paper, we identify Hurwitz Unstable Matrices directly by simply examining the sign patterns within the given real square matrix without the need to carryout any numerical computations. This capability is very much similar to what is currently being accomplished in the ecology field where their techniques give specific sign patterns which always possess negative real part (complex conjugate) eigenvalues independent of magnitudes, which is known as the Qualitative Sign Stability (QLSS) theory. In this paper, inspired by this QLSS theory, which restrict themselves to only predator-prey type interactions (i.e (+,-) type off-diagonal elements), we extend those concepts to engineering mathematical matrices which include all types of interactions, not just predator-prey type. Thus we combine predator-prey type interactions with all other interactions such as mutualism (i.e (+,+) type off-diagonal elements), competition (i.e. (-,-) type off-diagonal elements), ammensalism (i.e. (-,0) type off-diagonal elements and commensalism (i.e. (+,0) type off-diagonal elements) and identify the sign patterns which possess Hurwitz Instability (non-negative real part eigenvalues). This is accomplished by first creating a very large, un-bounded, open set of quantitative matrices, which are labeled as QuanTitative Sign, QTS matrices. Thus by combining the lessons learnt from QLSS Theory and examining the set of QTS matrices, together, we identify Hurwitz Unstable matrices directly from the sign patterns. We label those matrices as Magnitude Independent Sign Hurwitz Unstable, MISHU matrices. Such capability, naturally leads to significant computational savings and opens doors to speeding up computations in many engineering disciplines that get benefited by fast computation. Several examples are provided to illustrate this attractive capability.

Keywords: Machine Learning in modeling, estimation, and control, Modelling, Identification and Signal Processing, Healthcare systems
Abstract: Surface electromyography (EMG) provides a non-invasive and effective means for decoding upper-limb motion intent and is widely used in assistive robotics and rehabilitation systems. This study presents a comparative evaluation of multiple machine learning algorithms for EMG-based classification of upper-limb movements during passive and exoskeleton-driven motion. EMG data were acquired while the CLEVERarm, a custom-developed exoskeleton, executed three upper-limb movements crucial for rehabilitation: elbow flexion–extension, shoul￾der flexion–extension, and shoulder abduction–adduction. Continuous signals were segmented into 300-ms sliding windows with 80% overlap to utilize either raw temporal data for deep learning or extracted time-domain features for conventional algorithms, alongside rigorous compensations for class imbalance. Five classifiers, including Convolutional Neural Network (CNN), Support Vector Machine (SVM), k-Nearest Neighbors (kNN), Random Forest (RF), and Extreme Gradient Boosting (XGB), along with a weighted voting (WV) ensemble, were evaluated using leave-one-trial-out cross-validation. Results show the WV ensemble and XGB achieved higher overall performance compared to other algorithms, significantly outperforming non-tree-based classifiers such as CNN, SVM, and kNN (accuracy: 95% vs. 91%; F1-score: 95% vs. 91%; AUC: 0.99 vs. 0.95–0.98; p < 0.05). However, because the WV lacks statistical superiority over XGB, its computational burden makes XGB the preferred, streamlined model for real-world deployment. These findings demonstrate that robust decoding of passive motor intent is highly achievable using computationally efficient models like XGB. Clinically, this offers a vital step toward safe, low-latency, and patient-driven neurorehabilitation protocols for stroke and spinal cord injury survivors.

Keywords: Modeling and Validation, Modelling, Identification and Signal Processing
Abstract: Cranes are commonly used in industrial and construction applications to move large, heavy loads. The dynamics of these suspended payloads are highly oscillatory and often have low damping. Significant prior work has investigated the dynamics and control of payload swing; however, less work has investigated the mitigation of payload twist. This work proposes a six-degree-of-freedom model for a payload suspended by a single-looping cable. The validity of the model is examined using grey box estimation and experimental free response measurements collected from a small-scale tower crane.

Keywords: Modelling and Control of Environmental Systems, Motion and Vibration Control, Estimation
Abstract: A model of an IEA 15-megawatt Reference Wind Turbine blade’s structural dynamics is developed, using a generalizable approach, to widen the capabilities of the emerging Control-oriented, Reconfigurable, Acausal, Floating Turbine Simulator (CRAFTS) platform. The blade is discretized into a series of multi-body elements whose parameters are derived from the known distributions of structural and aerodynamic properties along the span. Eigen-analysis of the discretized model’s linearization is performed to tune damping factors and, for consistency with Timoshenko beam theory, the edgewise stiffness. The resulting properties of the model’s first two flap-wise modes and first edgewise mode reflect the nominal species. The model is substituted into an existing turbine configuration from the CRAFTS library to verify its aeroelastic response profile against the established finite element blade models, e.g., OpenFAST, HAWC2. The model’s steady-state response for standard inflowing wind speeds is consistent with OpenFAST and HAWC2. A controller for regulating mechanical fatigue loads is introduced to the system, and its effects on the model’s response are observed with the primary interest being the blades’ vibration. Both the fatigue loads and oscillatory deflection of the blades are found to be significantly reduced because of the regulator’s presence in the closed-loop system.

Keywords: Modelling, Identification and Signal Processing, Stochastic Systems
Abstract: Comparing nonlinear dynamical systems directly from trajectory data remains challenging because finite-horizon trajectory representations are generally coordinate dependent. Motivated by recent behavioral subspace approaches for linear systems, this paper introduces a Koopman-behavioral distance for nonlinear stochastic systems using multi-rollout trajectory data. State measurements are lifted through a common observable dictionary, and dominant lifted behavioral subspaces are compared using Grassmannian distances without explicit Koopman operator identification. We establish coordinate invariance, connection with recently proposed linear behavioral distances under exact lifted closure, and robustness to approximate closure. Numerical examples demonstrate coordinate invariance and parameter discrimination.

Keywords: Power and Energy Systems