Keywords: Adaptive and Learning Systems
Abstract: Autonomous systems are expected to adapt to new tasks and environments with little prior data, operate within the constraints of physical laws, and satisfy rigorous specifications for safety and performance. Meeting these demands requires moving beyond purely data-driven learning to hybrid methods that integrate control-theoretic reasoning, physics-based models, and formal specifications with modern machine learning. This talk will present a control-oriented perspective on learning that enables autonomy at operationally relevant timescales. I will highlight recent results showing how embedding physical knowledge and structured representations into learning architectures yields dramatic gains in data efficiency, generalization, and verifiability. These methods support on-the-fly adaptation and provide pathways to performance guarantees, even when data are scarce and environments are uncertain. The talk will conclude with a broader outlook on how control, learning, and formal methods together can bring trustworthy, rapidly deployable autonomy within reach.

Keywords: Robotics, Control Design, Machine Learning in modeling, estimation, and control
Abstract: Under impact disturbances, the states of autonomous robots jump, which deteriorates the system performance and stability. The hybrid dynamics nature presents significant challenges in control system design for online, real-time control applications. Model-based adaptive and robust control, which deals with bounded and continuous disturbances, requires accurate system dynamics. The proposed machine learning-based, region of attraction (RoA) -guided control suppresses dynamics jump with both RoA-tunable control and optimal impulse control. Stability is guaranteed automatically. Results show that the proposed control can deal with impacts that are 50 times greater than the normal input.

Keywords: Robotics, Intelligent Autonomous Vehicles, Unmanned Ground and Aerial Vehicles
Abstract: Robust navigation, environmental understanding, and efficient map representations are essential for autonomous mobile robotic systems, especially in unstructured and semi-structured environments. Traditional SLAM methods face challenges with computational complexity, communication bandwidth constraints, and in environments with sparser structured features. This research presents Semantic Navigation and Localization (SNL)-SLAM and explores how different types of object descriptors impact performance and memory usage for robotic navigation. Our methodology combines open-vocabulary object detection (OVD) models for object recognition in unfamiliar environments with factor graph optimization and novel methods for ground filtering, fusion of image and LiDAR data, and rich object descriptors. We experimentally evaluate our approach using an integrated hardware and software system. Our results show that SNL-SLAM achieves accuracy comparable to state-of-art non-semantic methods and produces maps with a one to two order reduction in memory footprint. We also explore, via ablation studies, relationships between different environments and the relative effects of several elements of object-level information on overall navigation, map accuracy, and map size.

Keywords: Robotics, Machine Learning in modeling, estimation, and control, Nonlinear Control Systems
Abstract: We propose a Physics-Informed Neural Network (PINN) framework to solve the nonlinear matching condition - a high-dimensional, first-order partial differential equation (PDE) that forms the core bottleneck to real-time energy shaping control in underactuated locomotion. This PDE is analytically intractable and computationally expensive to approximate numerically previously. We introduce a tailored PINN model design and a new alpha-masking technique that enforces boundary conditions without requiring loss balancing. To our knowledge, this is the first PINN-based controller validated in closed-loop on a biped, achieving stable, energy-efficient gait generation and scalable control design.

Keywords: Robotics, Modelling, Identification and Signal Processing, Modeling and Validation
Abstract: Retinal surgery requires extreme precision due to constrained anatomical spaces in the human retina. To assist surgeons achieve this level of accuracy, the Improved Integrated Robotic Intraocular Snake (I2RIS) with dexterous capability has been developed. However, such flexible tendon-driven robots often suffer from hysteresis problems, which significantly challenges precise control and positioning. In particular, we observed multi-stage hysteresis phenomena in the small-scale I2RIS. In this paper, we propose an Extended Generalized Prandtl-Ishlinskii (EGPI) model to increase the fitting accuracy of the hysteresis. The model incorporates a novel switching mechanism that enables it to describe multi-stage hysteresis in the regions of monotonic input. Experimental validation on I2RIS data demonstrates that the EGPI model outperforms the conventional Generalized Prandtl–Ishlinskii (GPI) model in terms of RMSE, NRMSE, and MAE across multiple motor input directions. The EGPI model in our study highlights the potential in modeling multi-stage hysteresis in minimally invasive flexible robots.

Keywords: Robotics, Nonlinear Control Systems, Discrete Event Dynamic Systems
Abstract: Time-delay embedding is a technique that uses snapshots of state history over time to build a linear state space model of a nonlinear smooth system. We demonstrate that periodic non-smooth or hybrid system can also be modeled as a linear state space system using this approach as long as its behavior is consistent in modes and timings. We extended time-delay embeddings to generate a linear model of two periodic hybrid systems—the bouncing pendulum and the simplest walker—with control inputs. This leads to a novel state history augmented linear quadratic regulator (LQR) which uses current and past state history for feedback control. Example code can be found at https://github.com/Chun-MingYang/koopman-timeDelay-lqr.git

Keywords: Robotics, Soft Robotics, Modeling and Validation
Abstract: This paper presents a unified framework to analyze the manipulability and compliance of modular soft-rigid hybrid robotic fingers. The approach applies to both hydraulic and pneumatic actuation systems. A Jacobian-based formulation maps actuator inputs to joint and task-space responses. Hydraulic actuators are modeled under incompressible assumptions, while pneumatic actuators are described using nonlinear pressure–volume relations. The framework enables consistent evaluation of manipulability ellipsoids and compliance matrices across actuation modes. We validate the analysis using two representative hands: DexCo (hydraulic) and Edgy-2 (pneumatic). Results highlight actuation-dependent trade-offs in dexterity and passive stiffness. These findings provide insights for structure-aware design and actuator selection in soft-rigid robotic fingers.

Keywords: Mechatronic Systems, Manufacturing Systems, Sensors and Actuators
Abstract: Additive manufacturing is a powerful tool for creating complex components that would be difficult to make using other techniques. In particular, multi-material 3D printing methods that incorporate electrically insulating materials and electrically conductive materials show promise for manufacturing systems with unique capabilities. In this work, a custom 3D printer that combines fused filament fabrication methods with direct ink write methods for silver nanoparticle paints is described. Using this machine, multiple thick layers of conductive traces can be generated in order to create high current linear and rotary electromagnetic actuators. Not only does this enable fully-3D-printed electromagnetic actuators, but it also enables single-print-session manufacturing of actuators with end effectors. Additionally, high torques, speeds, and motor efficiencies have also been observed, making these 3D-printed actuators competitive with traditionally wire wound motors. While thick traces are important for high current applications, fine pattern details with conductive feature sizes less than 0.5 mm are critical for high-frequency metamaterial radomes. To deposit the conductive paints accurately onto doubly-curved patterns, non-planar 3D printing methods are also developed. Combining these capabilities, the design and 3D printing of a two-conductive-layer tri-band frequency selective surface is described. Here, a large format sample is printed in a doubly-curved hyperbolic shape. Transmission and reflection measurements are captured and results show that the response between 5 and 30 GHz match well with the initial design. Overall, these examples illustrate the unique capabilities of multi-material 3D printing methods that use electrically conductive and electrically insulating materials. Not only can these techniques be used to create circuits, but they can also be used to manufacture high performance electromagnetic actuators and RF metamaterials in a unique automated fashion without additional assembly steps.

Keywords: Manufacturing Systems
Abstract: Digital Glass Forming (DGF) is a manufacturing process that uses a laser to heat glass, creating a workable zone that can be shaped. When a filament or fiber is continuously fed into the work zone, this method allows for the additive manufacturing of glass components. While is straightforward once the process reaches steady state, during unsteady periods it is difficult to achieve repeatable morphologies. Examples of unsteady deposition include starting and ending a track, sudden changes in deposition direction, and closing a connected track. In this talk, we will discuss our work in modeling unsteady glass deposition and the control techniques we have created to achieve precision deposition with use of a visual camera. Our real-time control system operates on Linux, which coordinates a fiber laser, a filament feeder, and a motion system upon which the part is fixed. A thermal camera, visual camera, and confocal displacement sensor are used for real-time data acquisition. All data is temporally and spatially registered. In this paper we consider the unsteady fabrication cases where deposition is starting and ending and where two tracks are being joined together. To achieve consistent starts, we are heating the filament to obtain a consistent tip size and then accelerating to the desired velocity. As the filament is heated, a ball forms at the tip and becomes larger as it moves up the filament. We will model the position and area of the ball as functions of laser power and filament displacement. We will use the model to control the size and location of the ball. Then, Iterative Learning Control (ILC) will be used to determine path and laser power profiles that provide constant starting and ending profiles in that, as measured by the camera, the track profile is consistent with the steady state track profile and starts/ends at the desired location. Next, we will use ILC to determine path and laser power trajectories to join two tracks. A set of experiments will be conducted using constant process parameters and another set of experiments will be conducted with control.

Keywords: Manufacturing Systems, Stochastic Systems, Path Planning and Motion Control
Abstract: Roll-to-roll (R2R) manufacturing is a continuous processing technology essential for scalable production of thin-film materials and printed electronics, but precise control remains challenging due to subsystem interactions, nonlinearities, and process disturbances. This paper proposes a Model Predictive Path Integral (MPPI) control formulation for R2R systems, leveraging a GPU-based Monte-Carlo sampling approach to efficiently approximate optimal controls online. Crucially, MPPI easily handles non-differentiable cost functions, enabling the incorporation of complex performance criteria relevant to advanced manufacturing processes. A case study is presented that demonstrates that MPPI significantly improves tension regulation performance compared to conventional model predictive control (MPC), highlighting its suitability for real-time control in advanced manufacturing.

Keywords: Path Planning and Motion Control, Adaptive and Learning Systems, Manufacturing Systems
Abstract: Machining often involves cutting parts of similar geometry repeatedly. To enhance productivity, quality, and cost effectiveness, the servo errors of the machine tool can be compensated via feedforward control, while the feedrate can be accordingly optimized to prevent tolerance violation or tool breakage. The repeated machining further provides opportunities for learning-based improvements. In this abstract, a feedrate optimization framework that is based on a hybrid model combining physics knowledge and part-to-part learning is proposed. In particular, the knowledge-based model is constructed based on a fixed structure with parameters to be tuned and is used for servo error pre-compensation. On the other hand, the part-to-part learning model is based on Bayesian linear regression trained on prior machined parts, where the confidence of the model increases as more confirmatory data are collected. Both models constitute the final prediction of the machine response and are both applied to feedrate optimization under kinematic, contour error and cutting force constraints. With online training, the iterative feedrate optimization permits progressively higher feedrates from one part to another, thanks to the increasing model confidence. Case studies are performed to demonstrate the effectiveness of the proposed methodology, showcasing the potentials of servo error pre-compensation and part-to-part learning for achieving faster machining.

Keywords: Manufacturing Systems, Control Applications
Abstract: Heat accumulation during laser powder bed fusion (L-PBF) of metallic parts can lead to undesired microstructure and mechanical properties. As the thermal history during the build process is highly dependent on the part geometry, there is a growing interest in developing geometric-dependent process parameters to improve build quality. Prior multiple studies investigated the control of laser power to mitigate heat accumulation in the build process. However, there can be limited control range during which the laser power can be adjusted, noting that there is a lower bound for the laser power below which lack-of-fusion will occur. In addition, such lower bound for laser power may vary depending on the initial temperature preceding scanning a new layer. This study will investigate the control of interlayer dwell time, which is the idle time (beyond recoating time) for additional cooling before starting to scan a new layer, either by itself or as a supplemental knob to the laser power control, for L-PBF fabrication of metallic parts. The control formulation is developed based on a part-scale finite-difference model that was previously validated experimentally. Numerical evaluation of thermal control with respect to interlayer dwell time is conducted through finite-element simulations. A square-canonical geometry of Inconel 718 from a past America Makes project, consisting of multiple thin walls and overhanging features, is used as a case study to demonstrate the effectiveness of the interlayer dwell control. Thermal-only and thermo-mechanical results will be presented to show the comparison among parts built under the default process parameters without any additional interlayer dwell time, controlled with a combination of laser power and interlayer dwell time, and controlled with pure interlayer dwell time.

Keywords: Machine Learning in modeling, estimation, and control, Adaptive and Learning Systems, Manufacturing Systems
Abstract: Selecting optimal process parameters in additive manufacturing (AM) is a challenge, given the number of variables that influence part quality and performance. From nozzle temperature and print speed to scanning sequence and toolpath planning, these parameters frequently interact in complex ways that often demand costly trial and error. This abstract presents a novel approach that leverages Bayesian optimization (BO) and Large Language Models (LLMs) to accelerate parameter selection by balancing exploration and exploitation. By drawing on the contextual understanding and domain knowledge of LLMs, the proposed method provides a head start to BO when observations are scarce, transitioning to classical BO for stronger statistical insight as more data becomes available. This synergy enables a more efficient exploration–exploitation trade-off, reducing the number of experimental trials required to reach optimal or near-optimal outcomes. The effectiveness of the method is demonstrated by tuning process parameters in material extrusion AM to minimize stringing, a common defect that occurs when filament oozes from the nozzle during non-printing moves, leaving unwanted strands on the final part. Experimental results show that the LLM-assisted BO approach requires fewer iterations than existing methods, achieving similar or improved print quality. The proposed strategy also reduces the time associated with trial-and-error experimentation and showcases its potential for broader applications of LLM-augmented optimization across diverse AM processes.

Keywords: Human-Machine and Human-Robot Systems, Adaptive and Learning Systems, Cognition modeling
Abstract: Control-based adaptive interventions are promising tools to promote equitable psychomotor skill development in classrooms. This work develops a novel multi-agent model tailored for such interventions, where students are represented as agents in a cooperative task involving simultaneous learning and goal achievement. The model’s low dimensionality facilitates validation, while its mathematical tractability enables analysis and control-based task design. Two students interacting with a ball-on-plate task are simulated to investigate the sensitivity of inequity and learning outcomes to diversity and collaboration. The results reveal that the outcomes depend nonlinearly on diversity and collaboration, suggesting the need for optimal design of learning tasks.

Keywords: Human-Machine and Human-Robot Systems, Assistive and Rehabilitation Robotics, Mechatronic Systems
Abstract: Estimating physical human-exoskeleton interaction forces is crucial for developing human-inspired control strategies in assistive devices. To improve our understanding of these interactions, in this paper, we present the design and integration of custom sensor modules to measure these forces and use a simulation approach to estimate them. The developed modules have force-sensitive resistors (FSRs) embedded in custom-designed casings attached to a compliant fabric thigh brace. The instrumented brace is integrated with an existing exoskeleton prototype to measure interaction loads between the user and the device. A dedicated calibration protocol and stationary human-in-the-loop tests were performed to validate the sensor response under various loading conditions mimicking exoskeleton assistance. Subsequently, the instrumented brace was used to record the interaction forces during a walking task performed by three subjects. The results highlight a correlation between interaction loads and exerted load or hip joint motions, as well as inter-subject variability due to large variations in thigh sizes that are an important ergonomic consideration. Finally, the brace was modeled in OpenSim using contact elements and simulated experimental conditions. The simulation results of the interaction forces at a selected contact point were validated by the experimental data.

Keywords: Human-Machine and Human-Robot Systems, Cognition modeling, Modeling and Control of Biotechnological Systems
Abstract: The determinants of human response to force perturbation include non-volitional mechanisms such as biomechanics and reflex responses as well as volitional means such as co-contraction and cognitive responses to haptic sensory feedback. Thus the modulation of neuromuscular admittance is receptive to instructions regarding these volitional strategies, even when the task remains constant. In this paper we investigate the influence of instruction on neuromuscular admittance when participants attempted to maintain the position of a manual control interface while experiencing unpredictable force perturbations. We found a clear distinction between trials where (N=10) participants were instructed to stiffen their arm by co-contracting or respond to forces they felt by producing opposing forces. Our findings have implications on the design of shared control strategies between humans and automation.

Keywords: Human-Machine and Human-Robot Systems, Mechatronic Systems, Assistive and Rehabilitation Robotics
Abstract: Rendering internal impedances is key for developing immersive experiences in virtual reality (VR). While previous research has primarily focused on the rendering of external impedances, rendering changes to the avatar remains an open question. Changes to the avatar are particularly salient for disease education and fostering empathy in healthcare professionals in conditions such as Rheumatoid Arthritis (RA). As a first step towards this immersive experience, this study explores whether humans can distinguish between joint and endpoint impedances. We developed a novel robotic system (Stiffness Emulation Robot) and a new controller for an existing haptics glove (HaptX G1), and conducted a binary discretion experiment which reveals that the majority of subjects can differentiate the two impedance types.

Keywords: Human-Machine and Human-Robot Systems, Robotics, Mechatronic Systems
Abstract: This paper presents the mechatronic development of a novel robotic bicycle platform that can both measure and influence human-bicycle dynamics using onboard sensors, a control moment gyroscope (CMG), and actuated steering. The standard Whipple bicycle model is first compared to a physics simulation in MuJoCo to demonstrate the suitability of the Whipple model for low-speed riding. A rider-in-the-loop data collection experiment is conducted, where the CMG is used to apply controlled roll-torque disturbances, and the rider-bicycle interaction forces are measured. The human rider's simultaneous control of steering and roll motion is modeled as a cascaded PID controller based on the functions of the peripheral and central nervous systems. The model is fitted to data for five subjects, and is demonstrated to stabilize the bicycle in a closed-loop. Our results capture the hierarchical structure of human bicycle control, similarities and differences between riders, and lay the groundwork for additional research in human-robot collaboration and rider-assistive technologies.

Keywords: Human-Machine and Human-Robot Systems, Unmanned Ground and Aerial Vehicles, Machine Learning in modeling, estimation, and control
Abstract: As the applications for Uninhabited Aerial Vehicles (UAVs) continue to grow, a quick and intuitive method of communication with these systems, independent of intermediate hardware, will be vital. Prior work has shown the feasibility of vision-based gesture recognition as a means of conveying commands to UAVs. However, practical considerations about gesture recognition performance from the dynamic viewpoints of UAVs have not been fully explored. In this work, we establish gesture set design criteria through a simulation-based study on static 2D (planar) and 3D (multi-planar) gesture sets. Data was captured using varying distance, angle, lighting, scene, and character models within the Unity game engine to construct a dataset of 147,420 images. Six instances of YOLOv11 object detection were trained on the data to capture the effects of these factors as image resolution is changed. The results of this study showed that designing gestures to have large and uniquely shaped bounding boxes leads to improvements in performance across a wide range of viewing angles that result from UAV operations.

Keywords: Machine Learning in modeling, estimation, and control, Adaptive and Learning Systems, Robotics
Abstract: Data-driven model predictive control (MPC) has demonstrated significant potential for improving robot control performance in the presence of model uncertainties. However, existing approaches often require extensive offline data collection and computationally intensive training, limiting their ability to adapt online. To address these challenges, this paper presents a fast online adaptive MPC framework that leverages neural networks integrated with Model-Agnostic Meta-Learning (MAML). Our approach focuses on few-shot adaptation of residual dynamics—capturing the discrepancy between nominal and true system behavior—using minimal online data and gradient steps. By embedding these meta-learned residual models into a computationally efficient L4CasADi-based MPC pipeline, the proposed method enables rapid model correction, enhances predictive accuracy, and improves real-time control performance. We validate the framework through simulation studies on a Cart-Pole system and a 2D quadrotor. Results show significant gains in adaptation speed and prediction accuracy over both nominal MPC and nominal MPC augmented with a freshly initialized neural network, underscoring the effectiveness of our approach for real-time adaptive robot control.

Keywords: Machine Learning in modeling, estimation, and control, Estimation
Abstract: Accurate state estimation requires careful consideration of uncertainty surrounding the process and measurement models; these characteristics are usually not well-known and need an experienced designer to select the covariance matrices. An error in the selection of covariance matrices could impact the accuracy of the estimation algorithm and may sometimes cause the filter to diverge. Identifying noise characteristics has long been a challenging problem due to uncertainty surrounding noise sources and difficulties in systematic noise modeling. Most existing approaches try to identify unknown covariance matrices through an optimization algorithm involving innovation sequences. In recent years, learning approaches have been utilized to determine the stochastic characteristics of process and measurement models. We present a learning-based methodology with different loss functions to identify noise characteristics and test the performance of these approaches in real-time.

Keywords: Machine Learning in modeling, estimation, and control, Estimation, Modeling and Validation
Abstract: This work presents a development and real-time implementation of an optimized Hierarchically Decomposed Graph Convolutional Network (HD-GCN) for intent classification of human motion based on camera-detected keypoints. To support both single- and multi-person scenarios, we adapted different pose estimation models for use in network training and real-time inference. The approach was validated using the NTU120 dataset, a widely adopted benchmark for action recognition that includes a broad range of day-to-day actions, as well as custom data collected in our lab representing actions related to industrial partners’ use-cases focusing on manufacturing related tasks. We implemented and compared multiple methods to achieve robust real-time intent recognition using only direct feed camera input to demonstrate the balance between computational efficiency and prediction accuracy

Keywords: Machine Learning in modeling, estimation, and control, Manufacturing Systems, Estimation
Abstract: Despite the potential of Virtual Metrology (VM), integrated frameworks combining standardized data handling, automated model development, and uncertainty quantification remain rare. This paper presents a scalable VM architecture that leverages Asset Administration Shells (AAS) for data integration, AutoML for modeling, and a practical UQ approach. We propose a novel but practically applicable method that connects GUM principles with ML-based uncertainty, aiming to support informed architectural decisions and foster robust, interpretable, and scalable VM deployment in industrial environments.

Keywords: Machine Learning in modeling, estimation, and control, Mechatronic Systems, Nonlinear Control Systems
Abstract: Model Predictive Control (MPC) faces computational demands and performance degradation from model inaccuracies. We propose two architectures combining Neural Network-approximated MPC (NNMPC) with Reinforcement Learning (RL). The first, Warm Start RL, initializes the RL actor with pre-trained NNMPC weights. The second, RLMPC, uses RL to generate corrective residuals for NNMPC outputs. We introduce a downsampling method reducing NNMPC input dimensions while maintaining performance. Evaluated on a rotary inverted pendulum, both architectures demonstrate runtime reductions exceeding 99% compared to traditional MPC while improving tracking performance under model uncertainties, with RL+MPC achieving 11-40% cost reduction depending on reference amplitude.

Keywords: Machine Learning in modeling, estimation, and control, Modelling and Control of Environmental Systems, Estimation
Abstract: Accurate wildfire detection and monitoring are essential for environmental protection. The fire front, a key indicator of fire behavior, is challenging to extract due to smoke and background noise. This paper presents Segment Anything Model (SAM)-based Fire Front Detection (SFD), a fully automated, pixel-level method for fire front extraction from imagery of unmanned aerial systems (UAS). SFD outperforms traditional edge detection and deep learning models, achieving superior precision, F1-scores, and IoU with few training data. Its modular design enables adaptation to other boundary-related tasks. These results demonstrate SFD’s potential to enhance wildfire analysis, offering a robust solution for improved fire monitoring and response efforts.

Keywords: Nonlinear Control Systems
Abstract: Hassan Khalil’s journey with control systems started half a century ago while doing his PhD at the University of Illinois under the supervision of Petar Kokotovic, from whom he learned the tools of singular perturbation. In this talk he will go over the main milestones of this journey that led to his work on nonlinear high-gain observers.

Keywords: Adaptive and Learning Systems, Manufacturing Systems, Optimal Control

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

Keywords: Control Applications, Machine Learning in modeling, estimation, and control, Manufacturing Systems

Keywords: Control Applications, Manufacturing Systems, Optimal Control

Keywords: Control Design

Keywords: Human-Machine and Human-Robot Systems, Manufacturing Systems, Adaptive and Learning Systems

Keywords: Machine Learning in modeling, estimation, and control, Adaptive and Learning Systems, Manufacturing Systems

Keywords: Machine Learning in modeling, estimation, and control, Manufacturing Systems, Cyber physical systems

Keywords: Manufacturing Systems

Keywords: Manufacturing Systems

Keywords: Manufacturing Systems, Control Applications

Keywords: Manufacturing Systems, Control Design, Optimal Control

Keywords: Manufacturing Systems, Robotics, Sensors and Actuators

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

Keywords: Mechatronic Systems, Manufacturing Systems, Sensors and Actuators

Keywords: Mechatronic Systems, Robotics, Manufacturing Systems

Keywords: Path Planning and Motion Control, Adaptive and Learning Systems, Manufacturing Systems

Keywords: Power and Energy Systems, Control Applications
Abstract: Gas Turbine (GT) engines are often chosen as the power source for islanded microgrids because their production flexibility and large capacity suit the needs of isolated systems, such as offshore platforms, hospitals, and refineries. Integrating batteries into GT-powered systems can improve operating cost, demand response, and overall reliability. Competing figures of merit, however, require appropriate control methods for system management. This paper examines the predictive control of a hybrid power plant consisting of a GT and battery, seeking to identify potential benefits of hybridization and advanced control over both short and long timescales, unified under a hierarchical control framework. A case study compares a baseline control approach to hierarchical predictive control in situations where both short- and long-timescale controller performance dominate. The proposed predictive control formulation is shown to support reliable power delivery by avoiding fault conditions while balancing among figures of merit associated with efficiency and degradation.

Keywords: Control Design, Large Scale Complex Systems, Machine Learning in modeling, estimation, and control
Abstract: Determining the optimal partitioning for distributed control is an NP-hard problem, making it computationally intractable for large-scale systems. This paper presents a learning-enhanced hierarchical branch-and-bound algorithm to find the near-optimal partitioning of a district heating network, which reduces the number of searched solutions while still finding an appropriate partition. The proposed method uses a bagged tree classifier to learn the critical communication links that lead to non-convergence of the non-cooperative communication-based distributed controller. The learned model achieves 91% classification accuracy in predicting promising branches. When integrated into the Branch and Bound algorithm, it identifies the same optimal partition as the exact approach while reducing the number of explored solutions by 88%.

Keywords: Estimation, Modelling, Identification and Signal Processing, Adaptive and Learning Systems
Abstract: Accurately estimating decision boundaries in black box systems is critical when ensuring safety, quality, and feasibility in real-world applications. However, existing methods iteratively refine boundary estimates by sampling in regions of uncertainty, without providing guarantees on the closeness to the decision boundary and also result in unnecessary exploration that is especially disadvantageous when evaluations are costly. This paper presents ε-Neighborhood Decision-Boundary Governed Estimation (EDGE), a sample efficient and function-agnostic algorithm that leverages the intermediate value theorem to estimate the location of the decision boundary of a black box binary classifier within a user-specified ε-neighborhood. To demonstrate applicability, a case study is presented of an electric grid stability problem with uncertain renewable power injection. Evaluations are conducted on three test functions, where it is seen that the EDGE algorithm demonstrates superior sample efficiency and better boundary approximation than adaptive sampling techniques and grid-based searches.

Keywords: Control Applications, Power and Energy Systems, Modelling, Identification and Signal Processing
Abstract: Thermochemical energy storage systems offer a sustainable solution for storing excess renewable energy. This work focuses on a particular application based on CaO/Ca(OH)2 for heat storage. To ensure safe and efficient hydration of CaO (storage discharging), a control strategy is required. A nonlinear controller design model is developed and refined, incorporating input delays and pump dynamics in the cooling water circuit. With experimental data, the parameters and input delays in the system model are identified. A model predictive control (MPC) strategy is then designed, with four objectives targeting reactor temperature, cooling water temperature and thermal output power. A proportional-integral-derivative (PID) controller is implemented for comparison. The control strategies are evaluated in terms of tracking performance, energy transfer and real-time feasibility. Simulation results show that MPC effectively tracks system temperatures and power while fulfilling input, output and state constraints. The results also confirm the real-time feasibility of the MPC approaches.

Keywords: Control of Smart Buildings and Microgrids, Nonlinear Control Systems, Power and Energy Systems
Abstract: This paper addresses the challenge of decentralized energy trading and load control in microgrids. Building on cooperative game theory, three Nash Bargaining (NB) games are formulated under limited, partial, and full information exchange to balance Pareto and social optimality while preserving Electric Vehicle (EV) user privacy. To ensure the charging station (CS) maintains its desired load profile, an overload penalty distribution (OPD) method is introduced to guide energy trading decisions. Recognizing the limitations of traditional alternating direction method of multipliers (ADMM) based methods in handling non-convex, large-scale problems, this paper proposes a distributed algorithm that decomposes the NB game into two tractable subproblems. The proposed framework enhances computational efficiency and scalability while ensuring fairness and coordination across participants in complex energy systems.

Keywords: Power and Energy Systems, Control Design, Modeling and Validation
Abstract: This paper presents the design and control of a solar photovoltaic (PV) converter system connected to a three-phase AC grid. The system operates in parallel with a local load, enabling power sharing between the PV system and the utility grid. An advanced Maximum Power Point Tracking (MPPT) algorithm, incorporating an emotional controller, ensures efficient power extraction and adaptive response to varying solar irradiance and load conditions. The proposed control strategy ensures maximum extraction of active power from the solar PV panel while simultaneously compensating for the reactive power demand of the AC load. This enables the three-phase AC source to operate at unity power factor. A Hardware-in-the-Loop (HIL) implementation validates the controller’s robust performance under varying load conditions and fluctuating solar irradiance.

Keywords: Machine Learning in modeling, estimation, and control, Manufacturing Systems, Cyber physical systems
Abstract: Machine learning plays a pivotal role in smart manufacturing by powering a wide array of decision-making tasks. As today’s manufacturing advances toward increasingly interconnected and data-rich ecosystems, the need for collaborative yet privacy-preserving machine learning is becoming more critical. Federated Learning (FL) has emerged as a promising paradigm that enables distributed model training across machines, systems, factories, or organizations without requiring the exchange of raw data. However, applying FL in industrial settings poses unique challenges, including non-IID data distributions, limited local data, and varying process configurations. This talk will showcase successful applications of FL in a range of manufacturing scenarios, including: (1) pixel-wise quality classification via semantic segmentation in additive manufacturing, (2) mixed fault diagnosis in rotating machinery, (3) dimensional accuracy prediction and part qualification in additive manufacturing, and (4) tool condition monitoring in ultrasonic metal welding. Furthermore, two new strategies for addressing data heterogeneity and non-IID challenges, including clustered FL and federated domain generalization, will be presented. Together, these studies demonstrate how FL enables cost-effective, privacy-preserving, and generalizable learning in manufacturing environments where centralized learning is impractical or undesirable.

Keywords: Manufacturing Systems
Abstract: Digital Glass Forming (DGF) is a heat assisted manufacturing process that uses a laser to create a workable zone that is subsequently shaped. The overwhelming majority of work in DGF has been in additive processes where glass is continuously fed via a filament or fiber into the work zone on a moving substrate to fabricate three dimensional parts. In these processes the shaping force is applied through the cold filament above the work zone. The ability of the cold filament to precisely shape the part decreases as the size of the work zone increases at elevated temperatures. To increase part morphological precision, after fabrication the part is locally heated and a tool is applied to the work zone to precisely shape the part. In this talk we will discuss our work in dynamic modeling and feedback control of the glass shaping process. We have created a control system operating in real time Linux that coordinates a fiber laser that heats the part and a motion system that moves the part relative to the stationary laser source and tool. A thermal camera, visual camera, confocal sensor, and three-axis force sensor are integrated into the control system for real time data acquisition that is temporally and spatially registered to the laser and motion system data. We conducted a series of experiments to model the shaping process force dynamics. During the experiments the visual camera determines the tool location relative to the part to determine the precise plunge depth, which will be less than the commanded plunge depth due to tool compliance. The input parameters to this dynamic model are tool velocity and local part surface temperature. After each shaping motion, a confocal sensor is used to measure the part profile. A feedback controller is designed to regulate the shaping force process. We conducted a second series of experimental studies to where the shaping force controller is implemented to fabricate a variety of shapes. The results are compared to experiments where constant process parameters are applied to shape the same parts.

Keywords: Manufacturing Systems, Control Design, Optimal Control
Abstract: This paper presents a control framework for microscale selective laser sintering (μ-SLS) that enhances geometric fidelity by predicting and regulating thermal distribution. Unlike the prior QP-form controller, the proposed LP-form method enforces stricter thermal constraints, generating laser masks that better match target geometries and reduce heat-affected zones (HAZs).

Keywords: Control Applications, Manufacturing Systems, Optimal Control
Abstract: Roll-to-roll (R2R) mechanical dry transfer is an enabling technology for high-throughput, environmentally friendly fabrication of advanced thin-film devices. However, precise control is required to ensure high-quality transfer, presenting a significant challenge due to nonlinear peeling dynamics, abrupt material property changes, and input constraints. This study proposes a sequential quadratic programming iterative learning control (SQP ILC) approach to regulate the R2R mechanical dry transfer process. The method leverages the system’s iterative structure to improve performance across successive transfer tasks while rigorously accounting for nonlinear dynamics and input constraints. Experimental validation on a lab-scale testbed and a case study on chemical vapor deposition (CVD)-grown graphene transfer show that SQP ILC significantly improves transfer quality with minimal online computation, making it a scalable solution for industrial R2R applications.

Keywords: Adaptive and Learning Systems, Modelling, Identification and Signal Processing, Manufacturing Systems
Abstract: Micro-additive manufacturing describes a broad domain in 3D printing used to fabricate high-resolution patterns for printed electronics, biosensors, and labels. Material jetting, a process in which ink is printed and interacts with the surface in liquid form, is commonly used to make printed electronics. Despite advantages in material diversity and drop-on-demand capabilities, deviations in the volumes of the printed droplets can lead to poor device performance. Real-time feedback control is often infeasible due to fast jetting dynamics and high-resolution feature sizes. In this work, we consider iterative methods to address limitations in real-time monitoring and control actuation. Iterative model updating in the form of a piecewise linear approximation and nonlinear curve fitting is used to derive updated models of the process to enable feedforward parameter selections for subsequent patterns. A comparison with traditional model-based ILC is incorporated and various error metrics are reported. Among the standard and incrementally variable experimental conditions, the results indicate that the proposed approaches offer faster convergence and a significant reduction in transient error when transitioning between different widths for the reference line.

Keywords: Adaptive and Learning Systems, Manufacturing Systems, Optimal Control
Abstract: Joining flexible structures such as in the assembly of fuselage sections in an aircraft and on-orbit assembly of modular space structures requires precise shape matching at the joint interface to avoid large local stresses that can lead to joint failure. However, it is difficult to accurately determine the forces needed (using analytical or numerical methods) for correcting the shape variations caused during the manufacturing process. Current assembly approaches use shims to reduce stresses in the presence of shape differences between the structures being joined. In contrast, the main contribution of this work is a data-based approach to experimentally reshape flexible structures to a desired shape. Specifically, data is used to develop predictive models that are then used to iteratively control the shape of those structures using a limited number of fixed actuators. The iterative control method accounts for uncertainties in the data-enabled predictive models and noise from the system to reshape the structure accurately. Experimental results with a relatively large (about 3 m long) structure demonstrate that the proposed approach accurately reshapes the structure with a substantial order-of-magnitude reduction (89%) in the maximum shape error.

Keywords: Multi-agent and Networked Systems, Adaptive and Learning Systems, Nonlinear Control Systems
Abstract: We propose a decentralized adaptive control strategy for heterogeneous multiagent systems in nonlinear Brunovsky form with ({pd})-dimensional n^{text{th}}-order dynamics where switching in communication topologies are time-triggered. Employing potential functions to ensure collision and obstacle avoidance, and using neural network-based estimation to account for uncertainties in the dynamics and disturbances of each agent, we establish decentralized control laws together with adaptive tuning laws and dwell-time requirements for transitions between communication graphs so that leader-following consensus is obtained. This integrated approach guarantees synchronized formations, stable performance, and effective disturbance rejection in evolving network configurations. Numerical simulations are provided to illustrate the results.

Keywords: Multi-agent and Networked Systems, Linear Control Systems, Intelligent Autonomous Vehicles
Abstract: Achieving consensus in multi-agent systems (MAS) under large communication delays remains a fundamental challenge, particularly in teleoperated leader-follower frameworks. This paper presents a passivity-based approach for compensating large delays in both operator-to-leader and agent-to-agent communications. The nominal delay-free communication architecture is shown in Fig. 1, where Fig. 1(a) illustrates the operator-to-leader link and Fig. 1(b) shows agen-to-agent communication. Under significant delays, the passivity-based transformation modifies communication structure as shown in Fig. 2. Unlike conventional consensus protocols that destabilize under large delays, the proposed framework ensures system-wide passivity and stability beyond classical bounds. A delay differential equation model is developed to derive delay-dependent stability conditions, showing robust convergence even when these bounds are exceeded. Simulations demonstrate instability in standard protocols [Fig. 3 (left)], while our method restores stability [Fig. 3 (right)]. Experimental validation using 3 physical mobile robots and 97 virtual agents (Fig. 4) confirms the results and performance in a real-world scenario. Compared to predictor-based schemes, our method offers superior delay tolerance, scalability, and ease of implementation. A video of the experimental results is available at url{https://youtu.be/pNbZpim5xf0}.

Keywords: Multi-agent and Networked Systems, Intelligent Autonomous Vehicles, Path Planning and Motion Control
Abstract: This paper tackles the problem of safe and efficient area coverage using a multi-agent system operating in environments with obstacles. Applications such as environmental monitoring and search and rescue require robot swarms to cover large domains under resource constraints, making both coverage efficiency and safety essential. To address the efficiency aspect, we adopt the Density-Driven Control (D2C) framework, which uses optimal transport theory to steer agents according to a reference distribution that encodes spatial coverage priorities. To ensure safety, we incorporate Control Barrier Functions (CBFs) into the framework. While CBFs are commonly used for collision avoidance, we extend their applicability by introducing obstacle-specific formulations for both circular and rectangular shapes. In particular, we analytically derive a unit normal vector based on the agent's position relative to the nearest face of a rectangular obstacle, improving safety enforcement in environments with non-smooth boundaries. Additionally, a velocity-dependent term is incorporated into the CBF to enhance collision avoidance. Simulation results validate the proposed method by demonstrating smoother navigation near obstacles and more efficient area coverage than the existing method, while still ensuring collision-free operation.

Keywords: Multi-agent and Networked Systems, Intelligent Autonomous Vehicles, Optimal Control
Abstract: This paper investigates the convergence conditions of Density-based Predictive Control (DPC) for non-uniform area coverage. In large-scale real-world scenarios, such as search and rescue or environmental monitoring missions, efficient non-uniform multi-agent area coverage is essential, as uniform coverage fails to account for varying regional priorities and operational constraints. To address this, we propose a novel multi-agent density-based predictive control strategy, DPC, grounded in optimal transport (OT) theory, particularly the Wasserstein distance. Given a pre-constructed reference distribution representing priority regions, DPC ensures that agents dynamically allocate their coverage efforts by spending more time in high-priority or densely sampled areas, thereby achieving effective non-uniform coverage. We analyze the convergence conditions of DPC by formulating the contraction mapping problem in terms of the Wasserstein distance. Additionally, we derive the analytic optimal control law for the unconstrained case and propose a numerical optimization method for determining the optimal control law under input constraints. Comprehensive simulations were conducted on both first-order dynamic systems and a linearized quadrotor model under constrained and unconstrained conditions. The results demonstrate that when the proposed conditions are satisfied, the Wasserstein distance locally converges, and the agent trajectories closely match the non-uniform reference distribution in all cases.

Keywords: Multi-agent and Networked Systems, Modeling and Validation, Motion and Vibration Control
Abstract: We investigate extreme event dynamics and allied synchronization in a heterogenous network of three coupled Josephson junction oscillators. We find extreme events in the local minima of a lower damped oscillator during librational-to-rotational motion transitions and simultaneously in the local maxima of a medium-damped oscillator exhibiting rotational-to-librational transitions. The extremeness of the events is validated by the threshold approach and reinforced using a statistical fit with the generalized extreme value distribution. By limiting our network size to 3N, we are enabled to view characteristics of the entire network in a single 3-dimensional phase space, illuminating a relationship between extreme events and synchronization phenomenon.

Keywords: Multi-agent and Networked Systems, Path Planning and Motion Control, Control Applications
Abstract: In this paper, we address the multi-agent rendezvous problem using Particle Swarm Optimization (PSO). We developed a thermodynamic PSO (TPSO) algorithm, a distributed PSO algorithm where each agent functions as a particle within the problem space. Additionally, we augmented an obstacle avoidance algorithm to ensure rendezvous in an unknown environment. Our simulations employ a mutli-agent system navigating an environment with unknown obstacles, with each agent independently utilizing the TPSO algorithm for movement and communications between the agents determined by a randomized connectivity matrix. We present performance results using a TPSO approach to observe behaviors of multiple agents.

Keywords: Machine Learning in modeling, estimation, and control, Modelling, Identification and Signal Processing, Optimal Control
Abstract: This paper investigates a novel downsampling technique based on optimal transport for managing imbalanced datasets in classification tasks. Downsampling is crucial for reducing dataset size while preserving essential statistical properties, thereby improving both classification performance and computational efficiency. Existing methods, such as random downsampling, NearMiss, Tomek Links, and Edited Nearest Neighbor, often fail to optimally preserve data distribution.

To address this, we propose a Wasserstein distance-based downsampling method that formulates an optimization problem to minimize distributional distortion. By leveraging the Wasserstein distance to measure dissimilarity between probability distributions, the proposed approach ensures that the reduced dataset retains key structural information. Additionally, we analyze the method's computational complexity and its impact on geodesic structures in Wasserstein space, highlighting its theoretical advantages over conventional techniques.

Simulation results on synthetically generated imbalanced datasets demonstrate that the proposed method outperforms existing downsampling techniques across multiple evaluation metrics, offering an effective and scalable solution for class imbalance in classification problems.

Keywords: Machine Learning in modeling, estimation, and control, Multi-agent and Networked Systems, Linear Control Systems
Abstract: We propose an integrated learning-based approach to address the reference tracking problem using a group of agents with unknown dynamics. The problem is divided into two subproblems: parameter identification and cooperative tracking. A parameter identification method is employed to estimate the dynamic of each agent using state and input measurements without the requirement for persistent excitation. The cooperative tracking problem is then formulated as an optimization problem with a defined loss function composed of tracking error and consensus error. Deep neural networks (DNNs) are used to produce optimal control policies based on state measurement and the reference signal. The weights of the DNNs are updated using a proposed gradient algorithm for dynamic systems to minimize the defined loss function. To enable continuous and in-time parameter identification, we propose an iterative algorithm that first carries out parameter identification followed by reference tracking capabilities. Simulation results demonstrate that given time-varying and unknown agent dynamics, the proposed algorithm still achieves parameter identification and reference tracking to ensure robust and adaptive tracking capabilities under complex environments.

Keywords: Machine Learning in modeling, estimation, and control, Optimal Control, Unmanned Ground and Aerial Vehicles
Abstract: Optimal trajectories that minimize a user-defined cost function in dynamic systems require the solution of a two-point boundary value problem. The optimization process yields an optimal control sequence that depends on the initial conditions and system parameters. However, the optimal sequence may result in undesirable behavior if the system's initial conditions and parameters are erroneous. This work presents a data-driven fuzzy controller synthesis framework that is guided by a time-optimal trajectory for multicopter tracking problems. In particular, we consider an aggressive maneuver consisting of a mid-air flip and generate a time-optimal trajectory by numerically solving the two-point boundary value problem. A fuzzy controller consisting of a stabilizing controller near hover conditions and an autoregressive moving average (ARMA) controller, trained to mimic the time-optimal aggressive trajectory, is constructed using the Takagi-Sugeno fuzzy framework.

Keywords: Machine Learning in modeling, estimation, and control, Power and Energy Systems, Cyber physical systems
Abstract: Cyber physical power systems (CPSs) have seen a considerable rise in malicious false data injection (FDI) attacks over the last decade. Metering infrastructure is most vulnerable to such attacks as they are spread out over a large topological area and their location often is accessible to consumers/generators. We consider such a scenario in our study where low magnitude stealthy FDI attacks are injected at different buses of an IEEE 14-bus, 30-bus an 118-bus systems. We propose a novel reduced sparse transformer (RST) neural network to detect the presence of FDI attack at multiple buses. The proposed RST uses a time series input of past measurements of active power at each bus received from the metering units and uses an encoder-only architecture to predict the presence of an attack at selected buses. We compare results with a baseline softmax or vanilla transformer neural network (TNN) and sparsemax attention-based TNN, which are the state-of-the art TNN architecture for time series, text and natural language prediction. The proposed RST architecture shows significant improvement in classification metrics for multi-label and single label for each of the attacked bus locations.

Keywords: Machine Learning in modeling, estimation, and control, Sensors and Actuators, Robotics
Abstract: Series Elastic Actuators (SEAs) are a promising alternative to conventional rigid actuation approaches to control robot manipulators performing dynamic tasks. In addition, the inherent compliance introduced with SEAs adds mechanical complexity that can be applied in robots that learn control actions through deep reinforcement learning. To investigate how the type of actuator drives dynamic task performance, we simulated a 2-link robot---both with SEAs and conventional rigid actuators---trained through proximal policy optimization to learn a cyclical task with contact. Using SEAs, the robot learned to actuate its joints in unique ways, which ultimately improved dynamic task performance.

Keywords: Machine Learning in modeling, estimation, and control, Underwater Vehicles
Abstract: Remote Operate Vehicles (ROVs) provide cost-effective solutions for underwater monitoring, inspection, and operations. However, achieving high-resolution state estimation remains challenging due to the difficulties in real-time positional data collection in underwater environments. This study proposes a Physics-Informed Neural Network-enhanced Error-state Kalman Filter (PINN-ESKF) to achieve robust and accurate position estimation by updating the covariance matrix in real time with the velocity reference generated by PINN. A physics-informed loss function, derived from the dynamic model of ROVs, is integrated into the training loop of a Deep Neural Network (DNN) to address the scarcity of high-quality real-world sample data. Simulations are conducted to evaluate the effectiveness of the proposed method, comparing the performance of PINN-ESKF against traditional EKF under low- and high-nonlinearity conditions. The results show that with high nonlinearities, the root mean square error is reduced by 88. 57% in the set-point scenario and 97. 56% in the path-following scenario.

Keywords: Automotive Systems, Control Applications, Estimation
Abstract: In a truck-trailer system, the overall system mass can vary significantly depending on how the trailer is loaded. This variation plays a crucial role in the performance of the Adaptive Cruise Control (ACC) system. Particularly, with a given braking capacity, the stopping distance can be largely affected by trailer loads. Thus, for safety and driver comfort, the ACC time gap setting can be modified when towing a heavy trailer. We investigate applications of estimation algorithms such as Recursive Least Squares (RLS) and Extended Kalman Filter (EKF) for trailer mass estimation using onboard sensors to enhance ACC system. By estimating the trailer mass, vehicle performance and safety can be improved during ACC operation. We demonstrate that the integration of such mass estimation approaches leads to an enhanced ACC system by providing smoother acceleration and deceleration, more consistent performance across available ACC time gap settings, and potential collision avoidance. Specifically, our contribution in this paper is the development of a time-gap adjustment mechanism for the ACC system based on the estimated trailer mass. This mechanism is supported by two components: an estimation activation method for improving the accuracy of trailer mass estimation, and a steady-state detection method for ensuring the estimation has converged before time gap adjustment. The efficacy of the approach is validated using IPG CarMaker simulation, which shows that by appropriately adjusting the ACC time gap, the system operates more effectively and safely across a wide range of trailer mass conditions.

Keywords: Automotive Systems, Control Applications, Intelligent Autonomous Vehicles
Abstract: This research introduces a new geometric controller for vehicle path tracking. Like the Stanley and pure pursuit controllers, it leverages geometric strategies to enhance lateral control in autonomous vehicles. Unlike conventional single-point tracking, where the controller targets a single future point on the path, this controller employs multi-point tracking, By analyzing various upcoming points along the vehicle's trajectory, this improves the car's perception and response to the path's curvature, with a Proportional-Integral (PI) controller used for longitudinal (speed) control. Modifying the look-ahead distances based on velocity and predefined ratio reduces initial steering errors and optimizes path alignment, especially during turns resulting in increased efficiency and safety in vehicle control systems.

Keywords: Automotive Systems, Control Design
Abstract: This paper is focused on the gearshift strategy of multi-speed electric vehicles (EVs) with multiple eDrives. With multiple eDrives, a supervisory controller is needed to coordinate torque inputs from different sources, as well as the clutch pressures, to achieve the desired shift quality. In this study, a model predictive controller (MPC) is developed for a vehicle with two motors, one at the front axle and the other at the two-speed rear axle. This MPC is hybrid due to the transitions between different gearshift stages. After linearization of the vehicle longitudinal dynamics, the formulated MPC becomes a quadratic programming problem and thus can be solved efficiently. Numerical studies show that the proposed method is promising for online implementation and is able to reach a balance between maintaining the desired acceleration and front-to-rear axle torque ratio during gearshifts. The proposed MPC framework can be easily expanded to other multi-speed EV powertrain structures in optimization-based approaches.

Keywords: Automotive Systems, Intelligent Autonomous Vehicles, Transportation Systems
Abstract: Battery electric vehicles (BEVs) face range limitations that pose significant challenges for freight transport, primarily due to limited energy storage, high energy demands, and variability in charging station availability, power capacity, and pricing. To overcome these challenges, a unified ecological driving (eco-driving) and economical charging (eco-charging) optimization strategy is proposed in this paper, where the main objective is to design the optimal vehicle speed profile and charging decision such that (a) mission-critical constraints---such as arrival time and final state of charge level---are satisfied, and (b) the total propulsion energy and charging costs are minimized. The proposed strategy supports both stationary and Dynamic Wireless Power Transfer (DWPT) charging systems. A Nonlinear Programming (NLP) formulation is presented to reduce computation time, and its performance is benchmarked against a brute-force method.

Keywords: Automotive Systems, Machine Learning in modeling, estimation, and control, Transportation Systems
Abstract: This paper proposes an innovative reinforcement learning (RL)-based autonomous braking algorithm that can be personalizable for optimal one-pedal driving (OPD) of electric vehicles. To address the shortcomings of OPD--including its counterintuitive braking, which confuses drivers, causes fatigue and discomfort, and promotes a lack of conformity/trust--we propose a framework integrates the state-of-the-art Twin Delayed Deep Deterministic Policy Gradient (TD3) RL agent with Learning from Human Demonstrations (LfD) via behavior cloning. An infusion term, lambda, controls the influence of human demonstrations on policy shaping, allowing varying levels of personalization. For the RL agent, a comprehensive reward function is designed to balance precise braking, human comfort, and regenerative braking energy. Seven unique agents with different lambda values are meticulously trained and evaluated against a baseline (lambda=0) and a human-like (HL) algorithm in a full-braking scenario. The results show that incorporating a moderate value of human demonstration (lambda=0.3) results in a more personalized and optimal control policy. Compared to the baseline, the proposed agent achieves an improvement of 212% in precise braking and 0.3% in energy recovery, and a reduction of 24% in root-mean-square (RMS) jerk and 10% in human-like action dissimilarity. In comparison to the HL algorithm, the proposed agent shows an improvement of 0.4% in energy recovery, and a reduction of 22% and 10% in RMS acceleration and jerk, respectively.

Keywords: Automotive Systems, Robotics, Intelligent Autonomous Vehicles
Abstract: An underactuated human driven vehicle is typically controlled by a minimal set of actuators available in the form of only one accelerator/brake pedal controlling the entire drivetrain, one steering wheel actuating a steering mechanism. Modern vehicles employ several active safety systems that the human driver does not control directly. A driver model closely mimics a human's driving behavior in closed-loop vehicle simulations, which are used to validate active safety systems such as ABS, TCS, etc. in a simulation setting. While driver models have been extensively researched for passenger cars, articulated vehicle driving involves specialized skills, thus making passenger car driver models unrepresentative for closed-loop vehicle truck-trailer vehicle simulations. In this paper, an articulated vehicle driver model is introduced that replicates a human driver's abilities at various levels of driving expertise.

Keywords: Unmanned Ground and Aerial Vehicles, Motion and Vibration Control, Robotics
Abstract: Drones tethered either to fixed ground stations or to a mobile vehicle, with the tether carrying power can potentially have very long flight times, that could be useful for applications in monitoring, surveillance or communication. Tethers are flexible, subject to varying tension, and capable of allowing vibrational modes that significantly impact drone motion. Tethered drone systems exhibit complex vibrational dynamics coupled with the rigid body motion of the drone on one end and the motion of a mobile ground platform at the other end. This paper presents a detailed dynamical analysis for a tethered drone system with the other end of tether fixed. The governing equations are derived and numerically solved using the finite element method. A modal analysis is conducted using the Stodola iteration method to determine the natural frequencies and mode shapes of the tether. Simulations demonstrate that varying lift forces alter the tether’s vibrational response, with lower lift forces exciting higher-order modes due to increased slack and varying natural frequencies, while higher lift forces stabilize the system due to absence of higher modes. The results highlight the necessity of considering tether dynamics in control strategies for tethered drone operations, particularly in applications requiring precise positioning and stability.

Keywords: Unmanned Ground and Aerial Vehicles, Optimal Control, Nonlinear Control Systems
Abstract: This work addresses the problem of setpoint tracking for tethered unmanned aerial vehicle (UAV) systems. Previous works have either used simplified models for the complex tension constraint to ensure a taut cable or proposed non-optimal strategies applicable to limited setpoints. The approach proposed here is based on the nonlinear model predictive control (nMPC), where the subtleties of implementing such a strategy for this nonlinear dynamic system with nonlinear constraints are addressed. We show faster tracking results than existing approaches that use the full-order nonlinear tension model. Additionally, we provide initial results for practical cases such as actuator failures, varying setpoints, and actuator saturation.

Keywords: Unmanned Ground and Aerial Vehicles, Path Planning and Motion Control, Intelligent Autonomous Vehicles
Abstract: This paper presents a reliable onboard intelligent system along with a keyboard-based control mechanism to simplify human-based remote beyond line-of-sight (BLOS) operation of Uncrewed Aerial Vehicles (UAVs) in unfamiliar environments with obstacles using Wi-Fi. While fully autonomous UAVs excel in open and outdoor spaces, human-operated UAVs are favored in customer-facing applications due to the pilot’s capacity for ethical decision-making, social adaptability, and innovative problem-solving in unexpected situations. However, learning to fly a UAV with a traditional Remote Controller (RC) has always been challenging since it requires adequate training and skills. BLOS operation and the communication latency (in both sending command signals and obtaining visual feedback) makes safe and collision-free RC operation of UAVs extremely challenging. The main contribution of this work is to provide a reliable assistive control mechanism that uses an onboard computer to assist the pilot in flying a UAV in an unknown environment using a keyboard. The onboard computer is responsible for maintaining stable flight by utilizing a 3D LiDAR to supply data to an SLAM algorithm, which creates a detailed 3D map of the environment and offers local position estimates in indoor or GPS-denied settings. Additionally, it ensures collision-free flights through a novel path planning and obstacle avoidance methodology using a combination of the Dynamic-Window Approach (DWA) and Artificial Potential Field (APF).We demonstrate via extensive simulations that, by implementing DWA and APF, we can help the operator maneuver the UAV safely and reliably towards the goal location, maintaining a safe distance from obstacles even in situations with communication delays.

Keywords: Unmanned Ground and Aerial Vehicles, Path Planning and Motion Control, Robotics
Abstract: Autonomous Mobile Robot (AMR) navigation in dynamic environments that may be GPS denied, without a-priori maps, is an unsolved problem with potential to improve humanity’s capabilities. Conventional modular methods are computationally inefficient, and require explicit feature extraction and engineering that inhibit generalization and deployment at scale. We present an Out-of-Distribution (OOD) Deep Reinforcement Learning (DRL) approach that includes functionality in unstructured terrain and dynamic obstacle avoidance capabilities. We leverage accelerated simulation training in a racetrack with a transition probability to parameterize spatial reasoning with intrinsic exploratory behavior, in a compact, computationally efficient Artificial Neural Network (ANN), which we transfer zero-shot with a reward component to mitigate differences between simulation and real-world physics. Our approach enables utility without a separate high-level planner or real-time cartography and utilizes a fraction of the computation resources of modular methods, enabling execution in a range of AMRs with different embedded computer payloads.

Keywords: Unmanned Ground and Aerial Vehicles, Robotics, Control Design
Abstract: This paper presents a new flocking control design that allows recoverable collisions among mobile agents. Existing flocking control methods typically include collision avoidance as one behavior rule in their control strategy design by ignoring practical and recoverable collisions in natural flocks (e.g., fishes, sheep, and ants). By modeling recoverable collisions through spring-damper interactions among agents, the proposed flocking control method can explicitly involve collision effects and agents' dimensions during flocking. Based on the collision model, a new Hamiltonian function is constructed to include collision energy in the flocking control design. Simulation results demonstrate that, despite the presence of collisions, the proposed new flocking framework enables agents to converge into a desired quasi alpha-lattice. Moreover, the Hamiltonian-based controller facilitates faster convergence and reduces overall energy dissipation compared to approaches that treat collisions solely as disturbances. This framework enhances the realism of flocking control and gives new insights into swarm robotics and collective movements.

Keywords: Unmanned Ground and Aerial Vehicles, Robotics, Power and Energy Systems
Abstract: This paper presents a control-oriented dynamic model, controller, and closedloop mobility characterization for the first wind-powered spherical rover capable of net upwind motion. This device, termed the Spherical Sailing Omnidirectional Rover (SSailOR), incorporates design features within a spherical, terrestrial rover that mimic the role that a centerboard (or keel) and lifting sails play in allowing net upwind motion for sailboats. Specifically, a traction hoop enables significant lateral resistance, thereby providing a nonholonomic constraint in the direction of travel. Lifting sails enable net thrust even when traveling significantly upwind, while also providing heading control. While providing unique capabilities, the SSailOR gives rise to a complex design and control space, where careful model-based design and control are necessary to ensure that the SSailOR can simultaneously (i) make net upwind progress, (ii) respond quickly to wind speed/direction changes, (iii) limit heel angle, and (iv) control its heading. To simultaneously address these challenges, we first present a controloriented dynamic model. This is followed with the presentation of a combined heading and heel angle controller. Finally, with the dynamic model and control structure in place, we present a detailed closed-loop Pareto analysis, which illustrates the tradeoff between transient and steadystate performance, along with the design features that favor one modality of performance over another.

Keywords: Manufacturing Systems, Robotics, Sensors and Actuators
Abstract: Quality inspection is a vital process in modern manufacturing to ensure product reliability. Robotic systems offer advantages for inspecting complex-shaped objects, but require precise spatial awareness to localize and orient targets within cluttered environments. To address this, we share our recent work on a learning-based Scene Point-Cloud Registration (iLSPR) framework for high-fidelity industrial scene reconstruction. iLSPR leverages point cloud representations and integrates three core components: a Multi-Feature Robust Point Matching Network (MF-RPMN), a Geometric-Primitive-based Data Generation (GPDG) method for efficient synthetic data creation, and a digital model library of target objects. During operation, vision sensors acquire point clouds of the scene. iLSPR identifies and registers the ground truth (GT) model of each object, digitally reconstructing the scene with high accuracy. MF-RPMN aligns point clouds using both raw and deep features, while GPDG enables effective pretraining. We evaluate our method on an Industrial Scene Object Pointcloud Registration (ISOPR) dataset that we newly created for benchmarking of accurate digital scene reconstruction. A real-world prototype using a Universal Robot UR5e confirms successful deployment of the system, reconstructing scenes with high fidelity.

Keywords: Human-Machine and Human-Robot Systems, Manufacturing Systems, Adaptive and Learning Systems
Abstract: Collaborative robots are increasingly deployed to work with human workers on disassembly tasks at remanufacturing sites. However, constrained workspaces, vision occlusions, and variable component conditions create highly uncertain disassembly environments, making fully autonomous manipulator motion planning challenging. Here we present a human-guided motion planning framework that leverages human force feedback to correct manipulator trajectories when the initially planned trajectory is unsafe or infeasible due to undetected obstacles or complex component removal tasks in disassembly. Specifically, we develop a torque-based optimization method to continuously detect human contact forces in real time and infer human intention when there is a physical contact between human and robot. Rather than relying on full demonstrations or manually dragging the robot, we design an online manipulator motion re-planner based on the inferred intention and the original trajectory to generate a new safe and feasible trajectory. Moreover, we develop an adaptive parameter-tuning mechanism that automatically aligns planning parameters to match operators' corrective behaviors, enhancing efficiency and reducing human effort. Experiments have been conducted on a 7-degree-of-freedom manipulator in a shared workspace, demonstrating that the proposed planning framework improves the efficiency and safety of human–robot collaborative disassembly while reducing human effort. The proposed planning framework can be further applied to enable a single operator to supervise multiple robots, offering a labor-saving solution for human–robot collaborative disassembly in intelligent remanufacturing.

Keywords: Mechatronic Systems, Robotics, Manufacturing Systems
Abstract: This work presents a mechatronic framework for robotic welding using a collaborative robot for automated execution of weld seams on metal assemblies. It addresses key challenges in robotic welding of parts that require many weld lines, including part registration, weld path planning, cobot trajectory generation and control, and weld process monitoring. A graphical user interface enables users to visualize the entire workflow and execute weld lines without the need for additional registration equipment. An integrated system is developed consisting of a UR10e cobot, a Miller Metal Inert Gas (MIG) welder, and a torch.

The workflow begins with part registration using three non-collinear points aligned with the physical part by the human operator guiding the cobot to each of the three points sequentially. These points are matched to the part CAD model and are used to locate the actual part in the robot base frame. Thus, the part is registered with respect to the robot base frame for weld path planning and corresponding robot trajectory generation. Once registered, the user selects the desired weld seams directly in the CAD model, and the system segments the weld path into continuous sections, and for each segment a sequence of Cartesian poses (position and orientation) is generated where the torch tilts at a 45-degree angle relative to the weld line.

To avoid collisions and ensure smooth transitions between weld sections, a safe plane is defined at a fixed height above the part. Between sections, the robot retracts to this safe plane to reposition and reorient before approaching the next weld segment. An infrared thermal camera provides real-time weld process monitoring and continuous temperature data during operation, enabling estimation of both the height and temperature of the weld pool; weld height is estimated using edge detection of the thermal profile.

The entire system is operated through a ROS-based GUI that handles registration, visualization in RViz, and real-time control of weld parameters. The GUI allows for weld selection and execution without manual programming, lowering the barrier for deploying in low-volume production.

We will discuss the framework and experiments in this presentation.

Keywords: Control Design
Abstract: Frog-legged robots are commonly used for silicon wafer handling in semiconductor manufacturing. However, their precision, speed and versatility are limited by vibration which varies with their position in the workspace. This work proposes a methodology for modeling the pose-dependent vibration of a frog-legged robot as a function of its changing inertia, and its experimentally-identified joint stiffness and damping. To further capture the wear of the robot components or the varying load caused by carrying or not carrying wafers, these parameters are further tuned by online data. The combined physics-based data-driven model is then used to design a feedforward tracking controller for compensating the pose-dependent vibration of the robot. The validation results show that the proposed method can effectively reduce the tracking error compared to a baseline time-invariant controller.

Keywords: Control Applications, Machine Learning in modeling, estimation, and control, Manufacturing Systems
Abstract: Craniomaxillofacial (CMF) fixation plates are widely used for reconstructive surgery to mechanically connect skeletal disunions due to accidents or diseases. In order to achieve geometric conformity between the fixation plates and skeletal structure of individual patients, incremental plate deformation involving manual bending and/or twisting has been traditionally practiced by surgeons. The process is time-consuming and requires years of training due to the nonlinear springback effect and limitations in visual inspection for quality control. This paper presents a method for autonomous plate deformation process control by means of multi-input-multi-output (MIMO) model predictive control (MPC). Prediction of the springback effect is based on the analytical formulations for plate bending and twisting to provide the physical basis for control formulation. A Gaussian Process (GP) model trained offline using Finite Element (FE) simulation compensates for errors due to simplification in the analytical deviations and serves as a surrogate for the State Space Model of the controller. Evaluation using emulated plate deformation experiments confirms the effectiveness of the GP-enhanced MIMO-MPC control method.

Keywords: Discrete Event Dynamic Systems, Large Scale Complex Systems, Path Planning and Motion Control
Abstract: The paper considers the problem of the safe operation of multiple agents with different capabilities and access authorities to effectively and safely accomplish a complex mission. This problem is decomposed into two main sub-problems. The first sub-problem is to obtain the desired configuration of the agent team so that the best coverage of a distributed target is achieved while distinct inaccessible regions are avoided. To achieve this, we first apply the principles of computational fluid dynamics to establish a nonsingular mapping between the motion space and a planning space that excludes all inaccessible regions. We then develop a novel deep neural network forward learning (DNNFL) to abstractly represent the target by a finite number of points specifying the desired configuration of the agent team. The second sub-problem is the mission planning that is defined as the event-triggered Markov Decision Process (ET-MDP) with constrained actions and components that are updated by a deterministic finite automaton.

Keywords: Modeling and Validation, Multi-agent and Networked Systems, Cyber physical systems
Abstract: System modeling frameworks can be categorized into imperative and declarative paradigms. A model's paradigm effects its efficacy: imperative models allow simple execution, while declarative models capture the behavior of the underlying system. This paper compares these paradigms, as well as functional and object-oriented frameworks, in light of physics-based systems. This is done by exploring the principles of systems modeling and simulation. Simulation is shown to be the composition of functions representing system behavior. Simulatable frameworks can be differentiated by their ability to identify and compose these functions for a specific input and output pairing. The various frameworks are explored, applying concepts more typically studied in computer science to general systems engineering. The frameworks are investigated by comparing simulations of a driven double pendulum in various modeling languages. Observations include that functional, declarative models allow for greater reusability and holistic system simulation.

Keywords: Modeling and Validation, Nonlinear Control Systems, Estimation
Abstract: The theory of chronological calculus, using sophisticated tools from differential geometry, has enabled higher-order averaging methods that are computable based on Lie-algebraic formulations. However, there seems to be some ambiguity in the literature regarding the class of systems on which higher-order averaging via chronological calculus can be applied to (dot{bm{x}}=epsilon bm{f}(bm{x},t) or dot{bm{x}}=epsilon bm{f}(bm{x},t;epsilon)). In this paper, we clarify the aforementioned ambiguity and clarify the conditions by which higher-order averaging methods using chronological calculus is applicable to systems in the form dot{bm{x}}=epsilon bm{f}(bm{x},t;epsilon). Additionally, we clarify the transfer of stability properties between an averaged system of finite order (e.g., first/second/third order-averaging) and the original nonlinear time-varying (periodic) system when the averaged system is asymptotically stable. We provide examples to illustrate the concepts discussed in this paper.

Keywords: Multi-agent and Networked Systems, Estimation, Stochastic Systems
Abstract: We present a framework to classify interactive states among bottlenose dolphins using data from a tagged individual and camera-based detections of others. A particle filter estimates continuous trajectories, while a Markov decision process infers states of association based on position, speed, and heading. Results reveal state-dependent speed differences and behavioral shifts over time. This approach enables fine-scale, data-driven analysis of social interactions in marine mammals.

Keywords: Stochastic Systems, Modeling and Validation, Nonlinear Control Systems
Abstract: Diffusive epidemic dynamics is typically analyzed assuming uniform disease spread, represented by isotropic diffusion operators in partial differential equation (PDE) models. However, anisotropic diffusion more accurately represents random epidemic spread biased by preferred directions, such as scenarios involving street grids, transport routes, and natural barriers. Investigating the effects of anisotropic diffusion using bifurcation analysis and numerical simulations, we find that directional mobility can lead to the emergence of new classes of self-organized epidemic spread patterns. Specifically, the results illustrate the effects of anisotropic diffusion on the onset of Hopf and Turing instabilities, as well as the consequent steady-state pattern formation. We present bifurcation diagrams that (1) establish the presence of Hopf and Turing bifurcations and (2) provide the contrast in the onset of Turing bifurcations between the isotropic and anisotropic diffusion cases. Additionally, studying the formation of steady-state Turing patterns, we find spatially symmetric and homogeneous equilibrium patterns in the isotropic diffusion case. However, the symmetry in the patterns is broken in the anisotropic cases, which yield striped patterns stretched in specific spatial directions dependent on the magnitudes of the anisotropic diffusion coefficients. In summary, the results uncover the role of anisotropic diffusion in triggering Hopf and Turing-type instabilities and self-organized pattern formation in a spatiotemporal, reaction-diffusion PDE epidemic model. The results are expected to be broadly significant beyond epidemic dynamics, since anisotropic diffusion generically represents the diffusive dynamics of collectives whose mobility is characterized by directional bias.

Keywords: Mechatronic Systems, Estimation, Modelling, Identification and Signal Processing
Abstract: This paper presents a novel diagnostic framework for electro-hydraulic systems based on adaptation and frequency domain analysis of a nonlinear inverse model. The framework comprises two sequential stages. First, a physics-informed nonlinear inverse model, structured as a generalized Wiener model, is estimated using an adaptive Kalman filter to ensure robust estimation under non-persistent excitation conditions. Second, inverse generalized frequency response functions, derived from the adaptive nonlinear inverse model, are continuously monitored to capture fault-induced behaviors in the frequency domain. Validation through multiphysics simulations and hardware-in-the-loop simulations demonstrates that the adaptive nonlinear inverse model accurately tracks system behaviors, and the inverse generalized frequency response functions exhibit selective sensitivity to different fault types and conditions, demonstrating the effectiveness of the developed diagnostic method for fault detection and isolation in electro-hydraulic systems.

Keywords: Mechatronic Systems, Estimation, Robotics
Abstract: This paper focuses on estimating the roll and pitch angles for a legged robot operating on a rough terrain. The system involves linear process dynamics and nonlinear output functions of multivariable (vector) arguments. An observer is developed for this class of systems where all the output nonlinear functions are assumed to be monotonic with bounded partial derivatives. In such cases, a single constant observer gain is shown to be sufficient for exponential convergence, and observer design LMIs for computation of the observer gain are presented. Furthermore, using switched gains, the developed observer can also be extended to non-monotonic systems. The legged robot application involves non-monotonic functions of vector arguments in the nonlinear measurement equations and also includes large unknown inputs. The use of the developed observer with an additional H_∞ disturbance rejection criteria clearly demonstrates the superior performance of the developed observer compared to traditional estimation techniques in experimental data.

Keywords: Motion and Vibration Control, Control Design, Modelling, Identification and Signal Processing
Abstract: This paper presents a numerically robust approach to multi-band disturbance rejection using an iterative Youla-Kucera parameterization technique. The proposed method offers precise control over frequency response shaping while maintaining numerical stability through a systematic design process. By implementing an iterative approach, we overcome traditional limitations in handling multiple frequency bands, explicitly accounting for performance trade-offs while minimizing undesired amplification effects. Numerical validation on a hard disk drive servo system demonstrates significant performance improvements, enabling enhanced positioning precision for increased storage density. The design methodology extends beyond storage systems to various high-precision control applications where multi-band disturbance rejection is critical.

Keywords: Modelling, Identification and Signal Processing, Motion and Vibration Control, Estimation
Abstract: Feedback control becomes fundamentally challenging when disturbances approach or exceed the sensor’s Nyquist frequency. We propose an information recovery method that reconstructs structured, aliased signals far beyond such Nyquist limitations using slowly sampled measurement data. This is achieved by a unique synergy of a multirate model predictor (MMP), the Internal Model Principle, multirate digital signal processing, along with finite impulse response (FIR) and infinite impulse response (IIR) designs. The effectiveness of the algorithm is validated through both numerical analysis and experimentation on a physical motor control hardware.

Keywords: Modelling, Identification and Signal Processing, Stochastic Systems, Linear Control Systems
Abstract: The design of filters seeks a separation of noise from a desired signal, and the boundary between both is a tradeoff that is a fundamental topic in signal theory. In the presence of signals wherein noise properties have time-varying components, this tradeoff is particularly challenging to optimize. The Mean Squared Error (MSE) is a standard performance metric for evaluating the performance of filters. For signals with non-white noise characteristics - which encompass nearly all real-world signals - the calculation of MSE typically requires repeated analysis across multiple experiments. Prior work by the authors introduced and extended Allan VARiance (AVAR) methods, which analyze variances within increasing data windows, to optimize Moving Average (MA) filters. That work suggested an equivalence between the time-consuming iterative process of using the MSE for filter optimization versus an analysis of the area under an AVAR curve, which can be calculated in one step.

This paper extends the use of the AVAR area method for selecting an optimal Finite Impulse Response (FIR) filter, where optimality is defined as the filter that minimizes the MSE between desired and filtered signals. Prior results are further extended to illustrate that the discrete integration of the AVAR curve yields a performance index that, in one step, generates the MSE-optimal filter for input with drift (random walk) corrupted by white noise. AVAR is compared against the MSE to show that both the performance indices give nearly equivalent optimal FIR filter designs. This AVAR FIR filter optimization is achieved with only one iteration versus hundreds of iterations to optimize filters using MSE calculations.

Keywords: Modelling, Identification and Signal Processing, Stochastic Systems, Linear Control Systems
Abstract: This work presents the application of Allan VARiance (AVAR) to determine the order of an FIR filter whose output approximates that of an IIR filter. AVAR methods are typically used to analyze the variance of static windowed averages of data; prior recent work by the authors extended AVAR methods to include optimization of moving average filter designs, and more recently optimization of finite impulse response (FIR) filters. One of the main advantages of FIR over IIR filters is that the output of FIR depends only on the data size equal to one more than the filter order, whereas the data size influencing the output of the IIR filter increases with the length of data. A consequence of this is that the AVAR optimization of IIR filters remains an unsolved problem, but one that is solvable if IIR filters can be well approximated by FIR filter designs. In this work, the similarity between FIR and IIR filters is quantified using the normalized distance between the AVAR curves of error. This approximation is demonstrated through time-domain results of a signal with both low- and high-frequency noise contributions, namely drift (random walk) input corrupted by white noise. The results show that the AVAR-equivalent FIR filter output is similar to that of the IIR filter output if one chooses a sufficiently high FIR filter order. In addition, an iterative algorithm is presented to quickly estimate the necessary order of an FIR filter that is AVAR-equivalent to an IIR filter.

Keywords: Control Design, Large Scale Complex Systems, Nonlinear Control Systems
Abstract: This paper presents a data-driven approach to designing decentralized Parallel Distributed Compensation (PDC) controllers for discrete-time Takagi–Sugeno (T–S) fuzzy large-scale systems. Instead of relying on system models, the proposed method uses measured system data, making it possible to design a decentralized PDC controller for T–S fuzzy large-scale systems even when modeling is unavailable. We show that if the collected data satisfy the specific rank condition, the T–S fuzzy large-scale system can be represented entirely using data. Based on this, we derive a data-driven formula for designing decentralized PDC state feedback controllers. The stability of the system is analyzed using quadratic Lyapunov functions, leading to a semi-definite programming (SDP) formulation that designs controllers by solving LMIs. The developed results are validated by coupled double-inverted pendulum systems.

Keywords: Estimation, Modeling and Validation, Transportation Systems
Abstract: Traffic systems encompass diverse driving behaviors, vehicle types, and infrastructure, making accurate modeling a challenge. While traditional car-following models capture some aspects of traffic dynamics, they struggle with the complexities of real-world interactions. To address this, we consider a hybrid traffic model by integrating two widely used car-following models to generate synthetic data. We then apply Dynamic Mode Decomposition with Control (DMDc) to estimate and predict its non-linear dynamics across different sampling rates, evaluating performance using root mean square error. The results demonstrate the effectiveness of DMDc in capturing complex traffic behavior, underscoring its potential to improve the understanding and prediction of real-world driving dynamics, ultimately advancing data-driven traffic modeling.

Keywords: Nonlinear Control Systems, Modelling, Identification and Signal Processing, Optimal Control
Abstract: This paper presents an output-feedback optimal tracking controller for a class of unknown nonlinear systems possessing full relative degree. The design procedure follows the standard Linear Quadratic Tracking (LQT) method using an approximate linear model of the system obtained via Koopman operator theory. A key contribution lies in identifying suitable observables for the Koopman method using only output measurements, thereby minimizing data collection costs and eliminating the need for full state information. We achieve this by recognizing that output derivatives serve as effective observables for this system class and employ a high-gain observer (HGO) to estimate these derivatives from output data. Overall, the proposed approach enables optimal control of the considered nonlinear systems without requiring prior knowledge of the system model. The controller synthesis and implementation rely solely on output measurements. This makes the proposed control strategy completely output-driven. We demonstrate the efficacy of the closed-loop system in controlling both an academic example and a power system with an infinite bus. Numerical comparisons with traditional linearization-based control highlight the performance benefits of the proposed Koopman-HGO approach.

Keywords: Robotics, Machine Learning in modeling, estimation, and control, Manufacturing Systems
Abstract: Defect detection is a crucial aspect of manufacturing and inspection processes. Traditional defect detection methods, including manual inspection and classical computer vision approaches, often struggle with scalability and adaptability to diverse types of defects. This paper presents an autonomous defect detection framework using a deep neural network tailored for 3D point cloud processing and optimized for edge deployment. Our approach combines high detection accuracy with lightweight point-based operation, making it suitable for industrial applications. We detail the data pipeline, the model architecture, and demonstrate the effectiveness of the network through experimental evaluation. The results show a strong performance in identifying defects efficiently, highlighting the potential of point-cloud-based deep learning for scalable quality control.

Keywords: Robotics, Machine Learning in modeling, estimation, and control, Modeling and Validation
Abstract: Autonomous manipulation for robots in unstructured environments is challenging due to the technical gap between perception and acting on the physical world. This challenge becomes even more difficult when an object's motion is constrained in space. This paper presents a method to rapidly estimate mechanical constraint models of systems in the environment using physics-informed deep neural networks (PINN). We develop a single network capable of estimating the motion of constrained mechanisms and present the methods for model training using synthetic data. By leveraging 6-dimensional constraints to selectively inform the loss function for different motions within the deep neural network, we achieve high accuracy in estimating the motions for linear and rotation constraints with sub-1 degree error, which is an order of magnitude improvement over recent studies. We experimentally evaluate our model on three real examples, validating the applicability of the approach.

Keywords: Robotics, Path Planning and Motion Control, Machine Learning in modeling, estimation, and control
Abstract: Autonomous robotic manipulation in unstructured environments faces many challenges and is hindered by capabilities that bridge the gap between perception and acting on the world. Action plans that are centric to object motion rather than end-of-arm tooling behavior may aid this. This paper presents an autonomous action planner for a feedback linearizeable system comprised of three base motions that can be leveraged on their own or in combination to give custom motion plans. The optimization routine for the three different types of motion are presented, which are integrated into physics informed neural networks. A component of this is the autonomy that decides how to use these plans independently and in combination. This approach is experimentally demonstrated on autonomous drawer opening, door knob turning, and door knob turning with door opening.

Keywords: Control Applications, Automotive Systems, Robotics
Abstract: This paper presents a robust, vision-based lane detection pipeline tailored for a scaled-down electric vehicle (DEFT) & traffic testbed that replicates real-world traffic scenarios. The proposed method systematically benchmarks three edge detection algorithms (Gabor, Canny and Sobel) and adopts the Sobel filter to balance detection accuracy with computational efficiency on embedded hardware. By integrating HSV-based color masking, morphological operations, and hyperbolic model fitting, the method achieves reliable lane characterization even under occlusions and discontinuities. Experimental validation of the proposed method shows smoother trajectories and improved yaw stability compared to predefined waypoint-based navigation, reducing significant lap time. Notably, the approach dynamically adapts to lane geometry without reliance on pre-mapped waypoints or external localization aids, supporting real-time deployment in resource-constrained environments.

Keywords: Estimation, Nonlinear Control Systems, Automotive Systems
Abstract: This paper presents vehicle lateral traversal velocity estimation method based on Lyapunov function and Unscented Kalman Filter (UKF) approach, and its application to One Touch Turn Signal (1TTS) feature enhancement. The vehicle lateral traversal velocity is estimated using vehicle dynamics signals acquired from the inertial measurement unit (IMU) and road line information acquired from the camera in the vehicle. This paper proposes a Lyapunov based method for initial estimation that is used as a propagation function for UKF to calculate the final vehicle lateral traversal velocity estimate. The estimation is used for automatic deactivation of the turn signals by detecting completion of a lane change, making the duration of turn signal operation adaptive to individual maneuver and scenario. The lateral traversal velocity estimation and the feature performance are verified using vehicle tests.

Keywords: Intelligent Autonomous Vehicles, Path Planning and Motion Control, Machine Learning in modeling, estimation, and control
Abstract: The use of artificial intelligence for planning safe and efficient trajectories for autonomous vehicles in dynamic and complex urban environments has grown rapidly. In this paper, we present a novel methodology for autonomous vehicle trajectory planning using Reinforcement Learning with a heuristic based reward function. The decision boundary formulated by a Support Vector Machine (SVM) classifier is used as a heuristic within the reward function to guide the agent along a smooth and collision-free trajectory toward a predefined goal position in a roundabout scenario. This heuristic based reward function is initially coupled with a Time to Collision (TTC) warning system, which is later replaced by dual SVM classifiers for trajectory and collision prediction of moving vehicles. The Soft Actor Critic (SAC) and Deep Deterministic Policy Gradient (DDPG) algorithms are used to train the agent to navigate safely through the dynamic roundabout scenario in minimum time to a goal position. The effectiveness of the proposed methodology is evaluated through simulations and compared against a Spatio-temporal lattice trajectory planner that uses an SVM based classifier as heuristic in its A* search algorithm.

Keywords: Large Scale Complex Systems, Transportation Systems, Path Planning and Motion Control
Abstract: Optimal movement of freight trains in railroad systems is vital for an energy-efficient, flexible, and sustainable transportation. This paper introduces a novel movement planning approach of freight trains based on the concept of virtual blocks as a step for autonomous freight railroad operation. The paper develops a train traffic model for a double track line, discusses the load dynamics, details a convex movement planning optimization model, and suggests a decentralized optimization procedure to solve the model. A simple application example inspired by an actual iron ore rail transportation scenario shows that the optimization model produces an optimal plan of realistic size in the range of seconds.

Keywords: Linear Control Systems, Control Design, Control Applications
Abstract: In this paper, we exclusively consider the special case of an nth order real square matrix A (n ≥ 2) and focus on how such real square matrix plays out its role in various linear algebra, matrix theory, and control theory situations, starting with a scalar system case. We emphasize and focus on one fundamental observation in matrix theory that every real square matrix of order n, can be expressed as the sum of few other matrices related to it, such as AT, Asym = (A + AT)/2 and Ask = (A − AT)/2. We show that A can also be viewed as the sum of 3 component real square matrices Amd, Amantd and Aict. If we approach the use of a real square matrix in all the matrix operations strictly under these facts, it turns out that, it can easily be shown that we do not need the determinant of that matrix, and thus do not need eigenvalue computation at all to assess the Convex Stability of a real square matrix A. This in turn implies that trace singularity becomes more critical than the determinantal singularity emphasized by the current literature linear algebra and matrix theory textbooks. The current literature Mapping theorem which states that the interior of the Unit circle maps into the Open Left Half of the Complex Plane becomes irrelevant for Convex Stability. In addition, it also implies that Mapping Theorem is also incorrect in eigenvalue based assessment of real state variable convergence.