Keywords: Machine Learning in modeling, estimation, and control

Keywords: Cyber physical systems, Power and Energy Systems, Control Applications
Abstract: Sustainability-targeting attacks (STA) are a growing threat to cyber-physical system (CPS)-based infrastructure, as sustainability goals become an integral part of CPS objectives. STA can be especially disruptive if it impacts the long-term sustainability cost of CPS, while its performance goals remain within acceptable parameters. Thus, in this work, we propose a general mathematical framework for modeling such stealthy STA and derive the feasibility conditions for generating a minimum-effort maximum-impact STA on a linear CPS using a max-min formulation. A gradient ascent descent algorithm is used to construct this attack policy with an added constraint on stealthiness. An illustrative example has been simulated to demonstrate the impact of the generated attack on the sustainability cost of the CPS.

Keywords: Intelligent Autonomous Vehicles, Power and Energy Systems, Unmanned Ground and Aerial Vehicles
Abstract: Autonomous mobile robot fleets must coordinate task allocation and charging under limited shared resources, yet most battery-aware planning methods address only a single robot. This paper extends degradation-cost-aware task planning to a multi-robot setting by jointly optimizing task assignment, service sequencing, optional charging decisions, charging-mode selection, and charger access while balancing degradation across the fleet. The formulation relies on reduced-form degradation proxies grounded in the empirical battery-aging literature, capturing both charging-mode-dependent wear and idle-state-of-charge-dependent aging; the bilinear idle-aging term is linearized through a disaggregated piecewise McCormick formulation. Tight big-M values derived from instance data strengthen the LP relaxation. To manage scalability, we propose a hierarchical matheuristic in which a fleet-level master problem coordinates assignments, routes, and charger usage, while robot-level subproblems, whose integer part decomposes into trivially small independent partition-selection problems, compute route-conditioned degradation schedules. Systematic experiments compare the proposed method against three baselines: a rule-based nearest-available dispatcher, an energy-aware formulation that enforces battery feasibility without modeling degradation, and an unlimited-charger formulation that models degradation but assumes a charger is always available, so that coordination of charger access is unnecessary.

Keywords: Machine Learning in modeling, estimation, and control, Modelling, Identification and Signal Processing, Estimation
Abstract: Accurate battery management under limited sensing requires models that infer internal electrochemical behavior from applied current and terminal voltage measurements. In practice, however, even physics-based models may exhibit systematic mismatch when important internal mechanisms are missed. This paper presents a physics-constrained identification framework for recovering latent dynamics in partially known lithium-ion battery models. The approach is demonstrated with a simplified case study, namely a reduced order li-ion cell model in which the electrode transport dynamics and related potential drop are assumed unknown. The unknown dynamics and related parameters are estimated using a novel formulation of the EM algorithm, where constraints are introduced in the M-step to enforce stability and physically admissible parameter bounds. Compared to standard constrained optimization, grey-box prediction-error estimation, and subspace-based black-box identification, the proposed framework achieves the smallest relative parameter error and the most accurate reconstruction of latent electrolyte states.

Keywords: Power and Energy Systems, Adaptive and Learning Systems, Machine Learning in modeling, estimation, and control
Abstract: Accurate identification of concurrent faults is essential for reliable solid oxide fuel cell (SOFC) operation, yet overlapping electrochemical, thermal, and flow signatures make fault isolation challenging, particularly given the limited availability of experimental studies. This paper presents a graph-attention-based deep learning framework for multi-fault classification using experimentally collected multivariate time-series data. The proposed architecture captures temporal dynamics and inter-sensor dependencies associated with coupled SOFC behavior. The model is evaluated on an 11-class task comprising five individual faults, five concurrent fault combinations, and one unknown class representing unforeseen operating conditions. It outperforms state-of-the-art baselines, achieving an accuracy of 0.97 and an F1-score of 0.94. The results demonstrate that graph-based sensor modeling improves the separability of interacting fault signatures, providing a practical framework for concurrent SOFC fault diagnosis and predictive maintenance.

Keywords: Power and Energy Systems, Control Design, Control Applications
Abstract: This work presents an Economic Nonlinear Model Predictive Control (ENMPC) framework to maximize the energy production of vertical-axis wind turbines (VAWTs). The formulation explicitly accounts for power losses and enforces constraints on rotor speed, thrust force, side-to-side force, and overturning moment. Unlike conventional regime-based controllers that rely on switching logic, the proposed ENMPC achieves seamless operation throughout the wind speed envelope using a single formulation with generator-torque-only actuation. Control performance is evaluated within the Offshore Wind/Water ENergy Simulator (OWENS), with fully coupled aerodynamic and structural degrees of freedom enabled. The results demonstrate consistent power maximization and robust constraint satisfaction, within acceptable tolerances, across a wide range of wind conditions with turbulence intensities prescribed according to International Electrotechnical Commission (IEC)} standard.

Keywords: Transportation Systems, Machine Learning in modeling, estimation, and control, Modeling and Validation
Abstract: Accurate estimation of battery health—specifically usable capacity, a key determinant of vehicle range—is essential for the successful operation and scheduling of battery electric buses (BEBs). Reliable state-of-health (SOH) assessment would enable transit agencies to plan service and maintain operations effectively. However, BEB performance and degradation are influenced by diverse and dynamic factors, including seasonal and regional climate variations, making SOH estimation from field data significantly more challenging than under controlled laboratory conditions. This study develops a long short-term memory (LSTM)–based machine learning framework to estimate the SOH of a fleet of 50 BEBs using weekly aggregated operational metrics. Because ground-truth SOH measurements are not directly available from the BEBs, the LSTM model is trained and tested on another fleet of battery electric commercial vehicles with similar usage patterns. A second-order equivalent circuit model (ECM), paired with an extended Kalman filter (EKF), is used to estimate battery SOH from this fleet. The comparison with the battery management system (BMS) information highlights a maximum root mean square error (RMSE) of 1.23%, confirming the good estimation capability of the EKF. Consequently, these EKF-based estimates are used to train, validate, and test the LSTM model on the same fleet before applying it to the target fleet of BEBs. The model shows good prediction performance and is specifically tested on the BEB with the highest mileage, achieving a maximum RMSE of 0.20% when compared to the onboard BMS measurements. By integrating the interpretability and physical grounding of the ECM–EKF framework with the scalability and data-efficiency of LSTM models, this approach provides a viable pathway for SOH estimation using low-dimensional, real-world operational data. The results demonstrate that aggregated metrics can support robust SOH prediction across fleets, enabling transit agencies to better understand range capabilities and degradation behavior under diverse operating conditions.

Keywords: Automotive Systems, Intelligent Autonomous Vehicles, Unmanned Ground and Aerial Vehicles
Abstract: Path tracking is a fundamental task in autonomous driving. Model predictive control (MPC) remains a dominant framework, but its online computational cost motivates learning-based alternatives. This paper presents a twin-delayed deep deterministic policy gradient (TD3) framework for vehicle path tracking that explicitly penalizes lateral acceleration and jerk, in addition to tracking error. To address the sampling inefficiency inherent in training on a high-fidelity vehicle dynamics model, an auxiliary training method—termed IBE (IIRS+BCR+EDT)—is introduced that integrates imitation-incentivized reward shaping (IIRS), behavioral cloning regularization (BCR), experience buffer pre-filling, and episodic demonstrator takeover (EDT) using a proportional–integral–derivative with feedforward (PIDF) controller as the expert. Simulation results show that the IBE-augmented agent converges with less than half the budget required by the baseline TD3, achieving lateral position errors within 0.4 m of the PIDF benchmark and reducing peak lateral acceleration by 24%.

Keywords: Automotive Systems, Modelling, Identification and Signal Processing, Modeling and Validation
Abstract: Accurate steering system modeling is essential for autonomous vehicle control design. This paper presents a physics-based steering actuator model using the bond-graph framework. Nonlinear friction in the rack-and-pinion mechanism is modeled using the Karnopp formula tion. Model parameters are identified through a two-stage process: structural parameters are estimated via frequency-domain fitting to a validated data-driven model, and Karnopp friction parameters are identified using vehicle-on-hoist step-response measurements. The proposed model is validated through frequency-domain and time-domain comparison with a previously developed data-driven steering model. Results show that the bond-graph model captures the dominant steering dynamics while providing a physically interpretable representation of the system.

Keywords: Estimation, Intelligent Autonomous Vehicles, Unmanned Ground and Aerial Vehicles
Abstract: While automating warehouse inventory processes is becoming increasingly important to create efficient storage solutions, inventory record inaccuracy (IRI) remains a challenge, motivating the deployment of robotic systems that can frequently, reliably, and autonomously check inventory. In particular, uncrewed aerial vehicles (UAVs) offer unique advantages as they can access inventory across varying shelf heights without additional infrastructure. We advance this development by introducing a tightly coupled framework of a fully autonomous UAV navigation system together with a fuzzy-logic driven inventory tracking management. Our navigation framework contributes a 3D LiDAR-inertial approach driven by an adaptive Kalman Filter to ensure high-frequency, minimal drift UAV localization. Further, the vision-based perception pipeline systematically detects inventory, employs a fuzzy classifier to extract specific product information, and accurately localizes items within the warehouse. Crucially, the system maintains track of the detected inventory by leveraging a Kalman Filter based tracking layer that utilizes Malahanobis distance and Chi-squared gating for track validation, alongside a Hungarian algorithm for robust data association. Additionally, the system detects and localizes any anomaly that it is trained to identify. The detection state is continuously integrated into a real-time Digital Twin (DT), which renders a high-fidelity replica of the UAV and warehouse layout populated with digital copies of products and anomalies detected.

Keywords: Optimal Control, Path Planning and Motion Control, Automotive Systems
Abstract: The Lyapunov optimization framework decomposes long-horizon optimization problems into a sequence of smaller, tractable subproblems. In autonomous vehicle control, the objective is to compute control actions that optimize a predefined performance metric such as travel time, energy consumption, tracking accuracy, or safety while satisfying strict constraints arising from vehicle dynamics, actuator limitations, and environmental factors, including obstacles. These constraints make real-time optimal control computationally challenging. Within this context, Lyapunov optimization can significantly reduce computational burden by transforming dynamic constraints into queue-based objective terms, while the associated Lyapunov function guarantees stability and convergence. Consequently, it provides an efficient framework for real-time optimal control in complex systems. In this work, we develop a novel optimal control approach based on Lyapunov optimization that achieves three times lower computational times while maintaining competitive tracking performance relative to conventional optimal control methods, thereby offering a principled alternative for real-time implementation.

Keywords: Path Planning and Motion Control, Intelligent Autonomous Vehicles, Automotive Systems
Abstract: Autonomous vehicle obstacle avoidance requires controllers that simultaneously respect vehicle dynamic limits, maintain passenger comfort, and ensure safety in critical situations. Conventional model predictive control (MPC) approaches typically rely on fixed vehicle models and constraints, which limits their adaptability to varying driving conditions and maneuver demands. This paper proposes a feasibility-driven parameter-varying nonlinear model predictive control framework for integrated lane keeping, obstacle avoidance, and emergency stopping. The vehicle dynamics are described using a nonlinear vehicle model with Pacejka tire forces to capture tire saturation during aggressive maneuvers. A single MPC controller is employed, while driving behaviors are adapted by coordinating vehicle model parameters and control constraints according to different driving modes, obtained from a previously developed physics-assisted unsupervised learning framework that characterizes distinct vehicle dynamic responses. A feasibility-based coordination strategy monitors obstacle avoidance constraints and automatically escalates control authority from comfort to sport mode when avoidance becomes infeasible, and finally to emergency stop when collision avoidance cannot be guaranteed. Simulation studies demonstrate that the proposed framework improves obstacle avoidance capability while preserving comfort under nominal conditions and ensuring safe operation in safety-critical scenarios.

Keywords: Unmanned Ground and Aerial Vehicles, Estimation, Automotive Systems
Abstract: High-speed autonomous racing platforms frequently experience Global Navigation Satellite System (GNSS) degradation or denial, leading to rapid error growth in an inertial navigator and challenging the overall Guidance Navigation and Control (GNC) stack. This paper presents a method for an optical flow-aided Inertial Navigation System Extended Kalman Filter (INSEKF) that incorporates body-frame velocity measurements while preserving sensitivity to vehicle slip and improving state observability. The system is validated by in the loop operation as the sole navigator for an autonomous Super Formula SF23 at Yas Marina Circuit. Results demonstrate robustness to extended outages, reduced attitude drift during dead reckoning, and improved orientation observability when coupled with GNSS without reliance on aggressive vehicle maneuvers. Offline testing shows that after a simulated 80 second outage, RMS position error is kept below 70 meters and RMS heading error is kept below 10 degrees. These results represent a roughly 2 order of magnitude performance increase in positioning and 1 order of magnitude performance increase in attitude when compared against a traditional inertial navigator.

Keywords: Optimal Control, Path Planning and Motion Control, Robotics
Abstract: Solving optimal control problems for high-dimensional mechanical systems subject to constraints remains challenging due to nonlinear dynamics and non-convex feasible sets. This paper introduces MAGMA(Multiplier–Augmented Geometric Motion Algorithm), an Affine Geometric Heat Flow (AGHF)–based trajectory optimizer that incorporates an augmented Lagrangian framework solved via a primal–dual flow. A theoretical characterization of the primal-dual AGHF is provided, establishing that MAGMA is a gradient flow on an appropriately defined energy. MAGMA is validated across a variety of challenging trajectory generation problems with obstacles, state limits, and actuation limits. Comparative evaluations against state-of-the-art solvers demonstrate that MAGMA achieves high quality solutions with significantly reduced tuning effort on the hardest problems. Code is available at https://github.com/roahmlab/MAGMA.

Keywords: Path Planning and Motion Control, Robotics
Abstract: Integrating legged robots into motion planning frameworks is challenging because navigation and balance goals must be met simultaneously. We present a hierarchical framework that utilizes task-based motion control to reduce legged dynamics to a unicycle model for motion planning. Then, using the standardized Benchmark for Autonomous Robot Navigation (BARN) environments, we benchmark three established local planners: Dynamic Window Approach (DWA), Potential Fields (PF), and Model Predictive Control (MPC) and validate these approaches via simulation and hardware experiments. Our results demonstrate that a hierarchical framework offers a robust, computationally efficient pathway for deploying quadrupedal robots in obstacle-dense environments.

Keywords: Path Planning and Motion Control, Uncertain Systems and Robust Control, Unmanned Ground and Aerial Vehicles
Abstract: Reachability-based Trajectory Design (RTD) is a provably safe, real-time trajectory planning framework that combines offline reachable-set computation with online trajectory optimization. However, standard RTD implementations suffer from two key limitations: conservatism induced by worst-case reachable-set overapproximations, and an inability to account for real-time disturbances during execution.

This paper presents RTD-RAX, a runtime-assurance extension of RTD that utilizes a non-conservative RTD formulation to rapidly generate goal-directed candidate trajectories, and utilizes mixed monotone reachability for fast, disturbance-aware online safety certification. When proposed trajectories fail safety certification under real-time uncertainty, a repair procedure finds nearby safe trajectories that preserve progress toward the goal while guaranteeing safety under real-time disturbances.

Keywords: Path Planning and Motion Control, Unmanned Ground and Aerial Vehicles, Agricultural Systems
Abstract: Crop-row centerlines provide geometric information for autonomous navigation, row-quality assessment, weed mapping, and crop-density and growth analysis, but their extraction from uncrewed aerial system imagery remains challenging for curved, discontinuous, and mixed row patterns. We present a training-free, rule-based framework that combines vegetation-support extraction, cell-wise orientation inference, wrapped-angle spatial regularization, multi-orientation reconstruction at structural transitions, spacing-guided recovery of missing segments, and inter-cell linking. The method was evaluated on georeferenced aerial imagery of corn and soybean fields containing straight and curved rows. Across all datasets, the reconstructed centerlines achieved a row overlap rate (ROR) > 99 % and a RMSE < 5 cm. By avoiding annotated training data and a global row-direction assumption, the framework provides a robust geometric representation for downstream field operations.

Keywords: Unmanned Ground and Aerial Vehicles, Path Planning and Motion Control, Optimal Control
Abstract: This paper analyzes the tradeoff between allocating samples for faster control frequencies or more-exact solutions when sampling-based model predictive control planners are used in extreme off-road mobility applications. Results from simulation show that increasing control frequency is more beneficial than increasing sample count by an equivalent amount. Implementations should target control frequencies of 50 Hz or higher, as the more-exact solutions found by high-sample-count optimizers can be detrimental to closed-loop performance when operating in highly-challenging scenarios at lower control frequencies. Physical trials using the simulation-suggested settings validate the planner’s ability to provide safe, extreme mobility in rugged off-road terrain.

Keywords: Unmanned Ground and Aerial Vehicles, Path Planning and Motion Control, Optimal Control
Abstract: Unmanned aerial vehicles (UAVs) can be used to track growing and moving boundaries such as those of wildfires where the boundary cannot be pre-specified. Toward this, we first present a model predictive control (MPC) formulation for this task, which systematically incorporates vehicle dynamics, evolving boundary models, and input constraints to enable precise tracking. While effective, solving the nonlinear optimization online incurs high computational cost, limiting real-time deployment. To address this, we propose a novel receding-horizon guidance law that replaces the optimization step with a closed-form solution based on steady-turn motion primitives embedded in a receding-horizon framework. This approach generates circular-arc trajectories in lieu of the computationally expensive optimization routine, while preserving the predictive nature of the formulation and enabling real-time onboard implementation. Simulation studies validate the method across varying UAV initial conditions, prediction horizons, and fire model parameters, demonstrating that it achieves tracking performance comparable to MPC while reducing computation time by several orders of magnitude.

Keywords: Healthcare systems, Modelling and Control of Biomedical Systems
Abstract: Stroke volume (SV) is a sensitive indicator of blood volume status but typically requires invasive measurement. Wearable-compatible physio-markers, such as the left ventricular ejection time/pre-ejection period ratio (LVET/PEP) and photoplethysmogram amplitude (APPG), have been shown to correlate with SV, although existing studies assume static relationships. In this work, we identified the dynamic relationships between SV and these physio-markers in hemorrhagic animals during resuscitation using a generative modeling approach. Results showed that an affine transfer function adequately describes the SV–physio-marker relationships. Compared with static models, the dynamic models significantly improved representation of the relationships. These findings demonstrate the dynamic nature of SV–physio-marker relationships and provide a foundation for developing improved non-invasive SV monitoring algorithms.

Keywords: Healthcare systems, Modelling and Control of Biomedical Systems, Machine Learning in modeling, estimation, and control
Abstract: This study focuses on developing Digital Twins (DTs) of glucose metabolism in people with Type 1 Diabetes (T1D) for personalized interventions. Our method combines physiology-informed mathematical modeling of glucose rate of appearance in blood after a meal intake, subcutaneous insulin absorption and glucose insulin kinetics, with a machine leaning based model recovery step from patient generated data, to map patients’ data to their virtual images. We use Liquid Time Constant Neural Network (LTC-NN) for model recovery, and data obtained from 10 in-silico patients simulated with a large scale metabolic model available in the literature. We evaluated replay fidelity under normal operating conditions as well as robustness to structured insulin and carbohydrate disturbances, against a state-of-the-art method for model twinning, namely Markov Chain Monte Carlo (MCMC). Our results indicate that the model parameters estimated with LTC-NNs consistently provided lower prediction errors than those identified with MCMC for all perturbation levels, paving the way for personalized interventions.

Keywords: Machine Learning in modeling, estimation, and control, Nonlinear Control Systems, Modelling and Control of Biomedical Systems
Abstract: Non-contact transport of small inertial particles in a viscous fluid can be achieved through the phenomenon of viscous streaming. Viscous streaming is the steady secondary flow generated by low-amplitude oscillations of a body immersed in such a fluid. In this paper, we develop control strategies that leverage a reinforcement learning algorithm together with human-designed heuristics to manipulate inertial particles in planar viscous streaming flows generated by circular cylinders oscillating transversely. We track particle motion within a rectangular planar region defined by four cylinders’ positions. This region serves as the state space and a discrete set of cylinder actuations serves as the action space. Our heuristic methods are compatible with more than one choice of learning algorithm; we focus on Deep Q-Network learning (DQN) in the present paper because of its broad familiarity. We use DQN to demonstrate three capabilities: (1) transporting multiple identical inertial particles to a common target, (2) separating two distinct inertial particles by size using differential response rates and topological barriers, and (3) sorting multiple inertial particles into size-based clusters. The central theme of our methods is the ability to shepherd multiple inertial particles simultaneously using an optimal policy learnt from a single-agent reinforcement learning algorithm. Particle dynamics follow the Maxey-Riley equation with Saffman lift and Fax'en correction terms. This work represents the first application of reinforcement learning to multi-particle transport and size-based separation in viscous streaming flows.

Keywords: Modelling and Control of Biomedical Systems, Healthcare systems, Control Applications
Abstract: Epilepsy affects more than 50 million people worldwide, where 30% of epileptic patients fail to respond to medication. Resective or ablative surgery can often reduce seizures for patients unresponsive to medication, but surgery success rates vary widely due to the difficulty in accurately locating the epileptogenic zone (EZ), which is the location where seizures originate in the brain. To confidently determine the EZ, patients undergo invasive electroencephalography (iEEG) and may wait up to three weeks in the hospital to capture spontaneous seizures. Our paper proposes a stimulation strategy to elicit seizures, which may lead to shorter hospital stays, more accurate EZ localization, and improved surgical success rates. We propose three methods for destabilizing fractional-order dynamical networks using minimal energy perturbation. Fractional dynamics have been shown to accurately represent neural behavior due to their ability to capture long-term memory. We simulate our approach on real iEEG data. Our results suggest that destabilizing fractional-order dynamical networks may provoke seizures. This work has the potential to improve EZ localization and surgical success rates for epileptic patients, enhancing their overall quality of life.

Keywords: Modelling, Identification and Signal Processing, Modelling and Control of Biomedical Systems, Mechatronic Systems
Abstract: We present a gastrointestinal (GI) phantom for in vitro motility measurement and an analysis framework for recovering propagation delay from multiple IMUs on deformable media without prior sensor alignment. Vertical- and in-plane-motion experiments emulate wave-like soft-tissue deformation. Three-axis gyroscope signals are projected using canonical maximization of coherence (caCOH), and the resulting signals are compared with planar-model, vision-based, and single-channel inertial references through cross-correlation. A sliding-window extension enables delay tracking and transition detection. Results show inter-sensor delay recovery in both motion conditions, tracking of imposed delay changes, and repeatable delay ordering under fixed-delay trials, supporting feasibility for deformable-motion sensing.

Keywords: Healthcare systems, Modelling and Control of Biomedical Systems, Control Applications
Abstract: This paper presents a prescribed performance control (PPC) approach to automated administration of intravenous anesthesia. PPC is an emerging non-parametric adaptive control approach that can enable persistent enforcement of performance specifications a priori at the control design stage in uncertain nonlinear dynamical systems. To cope with the confounding factors pertaining to intravenous anesthesia including control input saturation (infusion pump limitations and physiological constraints), measurement noise (patient monitor limitations), and external disturbances (surgical stimulations), we present practical extensions to robustify the conventional PPC: control law modification to soften the performance funnel when necessary, surrogate tracking filtering to enable smooth tracking error differentiation, and the synthesis of reference trajectory and feedforward input to enhance the tracking performance. Then, we show that the closed-loop system stability is preserved by the extended PPC against the confounding factors. The efficacy of the PPC approach was evaluated in simulation as well as hybrid hardware-in-the-loop plus simulated virtual patient (HIL+VP) testing based on a VP cohort generated from an established pharmacokinetic-pharmacodynamic model of propofol as the anesthetic. The results revealed remarkable reference tracking and disturbance rejection performance as well as consistency between simulation and HIL+VP testing.

Keywords: Human-Machine and Human-Robot Systems, Mechatronic Systems
Abstract: The work describes an overhead body-weight-support (BWS) system for in-home mobility assistance. The effort seeks to promote independent living for people with mobility impairments. With 12% of U.S. adults having a mobility disability with serious difficulty walking or climbing stairs, the proposed system has the potential to significantly improve in-home mobility, independent living, and quality of life. Note that overhead, cable-based, BWS systems have been successfully used by prosthetics, gait, and robotics researchers to avoid injury and damage during falls. BWS systems for fall protection have also been used when restraining subjects to perform daily activities such as walking or climbing stairs after traumatic injury. However, current BWS systems operate in controlled, obstacle-free spaces. Therefore, they are not suited for in-home mobility since they do not allow movement around (i) static obstacles such as tables, shelves, or structural columns and (ii) dynamic obstacles such as other people or objects, whose locations may vary. The current work adapts fixed BWS systems, such as industrial gantry systems with overhead cranes and lifting frames over hospital beds used to move patients, to enable mobility across multiple rooms. The proposed design consists of mobile trolleys that clamp onto roof-mounted beams and are connected to a harness with cables to support the human. Moreover, an output-sampling approach is presented that is coupled with inverse dynamics to find inputs that enable precision tracking of the selected force trajectories.

Keywords: Multi-agent and Networked Systems, Robotics
Abstract: This work considers string instability in constant-spacing policy (CSP) platoons without communication. It is well known that platoons will have string instability without communication with CSP if the vehicle dynamics has at least two poles at the origin. This work considers a modification of the vehicle dynamics, using input-output feedback linearization along with stable-pole zero cancellations, to retain only one pole at the origin. As a result, it is possible to design CSP that is string stable without communication. A disadvantage, with the reduction to only one pole at the origin, is that there is non-zero steady-state error for a constant platoon-speed. However, it is shown that a Pade-approximation-based, delayed-self-reinforcement (Pade-DSR) approach can substantially reduce the overall (transient and steady-state) spacing errors. Comparative simulations using a benchmark vehicle model show that the proposed approach avoids string instability and can reduce the overall error substantially (by 95%) when compared to the standard predecessor-following CSP platoons. Moreover, the resulting steady-state spacing deviation is more than 2 orders of magnitude less when compared to other string-stable approaches such as the constant-time-headway (CTH) policy.

Keywords: Linear Control Systems, Aerospace, Nonlinear Control Systems
Abstract: When controlling relative orbital motion using continuous or impulsive thrust guidance, calibrated feedback with respect to coordinates with reduced linearization error better represent full nonlinear Keplerian dynamics. The research presented here confirms previous findings that the accuracy of calibrated impulsive guidance improves the accuracy of linear dynamics and control methods. This work considers calibration of the time varying Linear Equations of Relative Motion (LERM) dynamics in the Cartesian Hill frame to analyze how it performs when used for impulsive and continuous thrust transfers. We find that the use of calibrated states can be advantageous in both transfer strategies depending on parameters such as transfer time and separation distance.

To form the calibrated continuous thrust control, we start with the familiar form of a Linear Time Invariant system. Next, we solve for the feedback gain by solving a LQR problem by minimization of the cost function assuming chosen positive semidefinite and positive definite R matrices. Instead of using the true state feedback, we apply the feedback to the calibrated state, where K minimizes the cost in the LQR problem. In addition, we may apply a Lyapunov Floquet transformation to extend control to the Linear Time Varying LERM system.

Given a desired final state and an initial state, the LERM STM allows one to compute the necessary delta-v in order to execute the maneuver in the prescribed time. The STM determines the necessary maneuver given initial (calibrated) state, target destination, and transfer time. After finding the prescribed maneuver according to a given guidance solution, we feed the maneuver into the nonlinear dynamics integrated in the inertial frame. The calibrated feedback arrives closer to the target than the true state feedback.

By Monte Carlo simulation, we find that calibration does not provide an appreciable benefit over uncalibrated impulsive guidance over short transfer times, but longer transfer times show that calibrated solutions result in better positional accuracy. We give the chief orbit elements and assume normally distributed deputy orbit element differences used to generate the Monte Carlo samples.

Keywords: Mechatronic Systems, Unmanned Ground and Aerial Vehicles, Path Planning and Motion Control
Abstract: Emergency obstacle avoidance by differential-drive mobile robots requires balancing short reaction distances against wheel slip and encoder-based heading error during aggressive braking or steering. In this experimental study, five avoidance maneuvers were tested at three approach speeds on a Pololu Romi 32U4 wheeled mobile robot platform. The speed controller used wheel encoder measurements, an onboard gyroscope recorded angular velocity, and an overhead camera used ARUCO markers to independently measure position and heading.

To quantify maneuver aggressiveness, we introduce a new parameter, which we refer to as the normalized rotational demand. It measures the maximum turning related acceleration commanded at either wheel and is normalized using the most aggressive spin tested on the robot and surface. Note that this reference is not a measured friction limit. Firth-penalized logistic regression was used to estimate the front-bumper distance at which the fitted probability of successful avoidance reached 90 percent. Across three camera-instrumented experimental scenarios, 1,050 trials were conducted, and the principal balanced, block-randomized campaign retained 539 of 540 trials. As speed increased from 45 to 65 centimeters per second, the absolute difference between encoder and camera measured heading changed by 6.8 to 7.0 degrees for the highest-demand spins, but by no more than 1.5 degrees for the lower-demand turns. The estimated distance required for 90 percent success increased by 3.8 to 5.7 centimeters for four maneuvers, while the smooth turn showed a nonmonotonic trend. Across 431 turning trials, greater heading disagreement was associated with failure after accounting for maneuver, speed, and reaction distance. This disagreement is consistent with rotational slip, but because slip was not independently varied, the experiments do not establish that slip caused the failures. Clearance variability also increased with speed.

In summary, the parameter normalized rotational demand identified the maneuvers most sensitive to speed, but maneuver selection requires speed specific tradeoffs among reaction distance, slip related heading error, and clearance variability.

Keywords: Adaptive and Learning Systems, Optimal Control, Linear Control Systems
Abstract: A model-based and data-driven reinforcement learning framework is developed for policy iteration (PI) and value iteration (VI) in optimal HVAC control. A continuous-time two-input thermal model with integral-state augmentation is considered under a common quadratic objective and identical initialization conditions. PI employs a complete Lyapunov-based policy evaluation, whereas VI performs successive Bellman value updates. Both approaches converge to the same optimal Riccati feedback solution, but exhibit markedly different learning characteristics. PI achieves substantially faster convergence with stabilizing initial policies, whereas VI converges more gradually and offers greater flexibility in policy initialization. The results reveal a fundamental tradeoff between convergence speed and initialization requirements in learning-based HVAC control.

Keywords: Modelling and Control of Biomedical Systems, Healthcare systems, Control Applications
Abstract: Overweight and obesity affect roughly 70% of adults in the U.S., contributing to many leading causes of death, including cardiovascular disease and diabetes. Standard behavioral interventions can achieve weight loss, but long-term maintenance is rare and influenced by a mix of psychosocial, behavioral, and physiological factors. This study introduces a semi-physical dynamic model with a control strategy to support FLASH (Finding Lasting Approaches to Sustaining Health), an ongoing intervention aimed at maintaining weight loss. In FLASH, participants who have lost at least 3% of their body weight engage in a 12-month system identification experiment featuring four smartphone-based interventions focusing on emotion, motivation, eating habits, and physical activity/sleep. Body weight and psychosocial factors are monitored daily. Our interdisciplinary team created a conceptual diagram linking psychosocial constructs to intervention components and, via energy balance, to weight change. This has been formulated into a fluid-analogy model adapted from supply chain management, which treats each construct as inventory with inflows and outflows, yielding a system of 12 differential equations that relate total energy intake and physical activity, along with a three-compartment energy balance model for body weight computation. Using this model, we formulate a control-based Just-in-Time Adaptive Intervention (JITAI) as a three-degree-of-freedom Kalman-filtered Hybrid Model Predictive Controller (3DoF-KF HMPC) that acts on a Mixed Logical Dynamical (MLD) representation of the FLASH system. These four intervention components are entered as binary decision variables, selected by the controller at weekly intervals, to track a weight setpoint and reject disturbances (e.g., perceived busyness, life events). The three degrees of freedom independently tune setpoint tracking, measured disturbance rejection, and unmeasured disturbance estimation responses, allowing speed of response to be shaped independently from robustness. A Kalman filter reconstructs the augmented behavioral and physiological state from a single measured output (body weight). Illustrative cases of its usefulness will be presented.

Keywords: Manufacturing Systems, Mechatronic Systems
Abstract: This work describes a differential flatness-based semiautonomous control algorithm to reduce payload swing and avoid collision while operating tower cranes. Tower cranes are widely used to transport heavy payloads in large, congested construction areas. During operation, an operator generally commands the crane while simultaneously monitoring the payload motion and swing, and trying to avoid surrounding obstacles. This is challenging because the payload is suspended by a flexible cable and is not directly actuated. The motion of the crane can therefore excite payload oscillations that persist even after the commanded motion has stopped. These oscillations reduce positioning accuracy and can increase the risk of collision with nearby structures. The current work considers a semi-autonomous tower crane system in which a human operator continues to specify the desired payload motion, while the controller assists with payload-swing reduction and collision avoidance. The main contributions of the current work are the following. (1) Semi-autonomous tower-crane teleoperation: A differential-flatness-based framework is developed in which the human operator specifies the desired payload motion while the controller generates the corresponding dynamically consistent tower-crane motion. (2) Integrated payload swing reduction and collision avoidance: Operator commands are modified when necessary to avoid obstacles and reduce payload swing, and smooth payload trajectories are generated so that collision-avoidance actions do not introduce unnecessary payload swing. (3) Real-time trajectory generation from operator commands: The desired payload trajectory is continuously updated from the operator’s velocity commands, allowing the method to operate without requiring a predefined payload path.

Keywords: Modeling and Validation, Soft Robotics, Robotics
Abstract: As humans expand into cislunar space and beyond, extravehicular and surface exploration will require astronauts to perform repetitive manual tasks under lunar, Martian, and other gravity fields. Wearable robots that assist musculoskeletal joints offer a promising lightweight, passive solution. However, designing them for widely varying gravity is uniquely challenging: optimizing exosuit parameters against gravity requires accurate models, which are scarce for soft, compliant systems, while ground testing is limited to 1g and true reduced-gravity facilities are costly and time-limited.

We propose emulating altered gravity for a test arm/exosuit by submerging it in water: tuning the displaced volume and added ballast mass on the link balances buoyant and gravitational forces to achieve an arbitrary effective gravity, from lunar and Martian levels to hypergravity. We derive an analytical expression for this effective gravity on a single-link revolute joint, yielding the volume and mass required to hit a target gravity. Using this expression, we optimize exosuit parameters directly at the target gravity, simulate the corresponding submerged system, and compare predicted motor torque and work between the two. The passive exosuit harnesses gravity itself: while lowering the arm, gravity stretches a spring via a cam-pulley mechanism, storing energy that is released to assist movement while raising the arm.

At low and microgravity levels, hydrodynamic drag and added fluid inertia become significant relative to the small, buoyancy-driven static torque, causing submerged and dry-optimized responses to diverge. We compensate for this discrepancy by estimating these hydrodynamic torques, scaled by a conservative estimation-quality factor, which restores close agreement between the two systems. At higher gravity levels, the systems show strong intrinsic agreement. These results validate submerged emulation as a practical, low-cost testbed for designing and analyzing wearable robots across multi-gravity environments—from the Moon and Mars to hypergravity worlds.

Keywords: Transportation Systems, Modeling and Validation, Control Applications
Abstract: This research presents an equivalent circuit modeling (ECM) framework for representing macroscopic traffic dynamics using concepts traditionally employed in electrical circuits. The proposed formulation establishes analogies between vehicle accumulation and electric charge, traffic flow and electric current, and the number of vehicles within a road segment and capacitor voltage. A road link is spatially discretized into interconnected cells represented by capacitors, while nonlinear controlled current sources govern vehicle transfer between adjacent cells according to upstream demand, downstream receiving capacity, and maximum link flow. An additional capacitor is introduced at the link entrance to explicitly represent vehicle queuing when the downstream road cannot accommodate the incoming demand. The resulting nonlinear ECM is shown to provide an equivalent circuit representation of a spatially discretized Lighthill–Whitham–Richards traffic model and can alternatively be interpreted as a nonlinear RC ladder network with traffic-dependent resistances. A steady-state and small-signal analysis is further developed to characterize the model under uncongested and congested operating regimes. The resulting linearized models provide an intuitive circuit-based interpretation of traffic-wave propagation, showing downstream propagation of disturbances under uncongested conditions and upstream propagation during congestion. The framework is additionally extended to merging and diverging traffic nodes through controllable current sources and optimization-based flow-allocation mechanisms that account for downstream capacity and prescribed merge or split priorities. Numerical comparisons with the Cell Transmission Model demonstrate that the proposed ECM captures the principal evolution of vehicle accumulation, congestion formation, dissipation, and input queuing, while offering a compact, interpretable low-fidelity representation suitable for analysis and traffic-control applications.

Keywords: Multi-agent and Networked Systems

Keywords: Manufacturing Systems, Mechatronic Systems, Control Applications
Abstract: In Digital Metal Forming (DMF) processes, a part is incrementally reshaped until it achieves the desired morphology and structural properties. At each increment the part is measured and the process plan for the subsequent step is generated. Adaptive process planning is needed due to the complexity of the forming process. To achieve this adaptive process planning, in-situ morphology measurements are required. In this talk we will discuss the use of morphology measurements in three DMF processes. In the first process, a plate is bent (or rebent) to a desired angle. The plate is displaced a small amount to learn a model, that is used for a single feedforward step and to compute the process gains in subsequent steps. Camera measurements are used to assess the process at each increment and a feedback controller is used to achieve precision bending even in the face of process uncertainties. In the second process, a plate is reshaped. In the first part of this process, the plate is stretched to a desired by iteratively sweeping a half sinusoidal pattern. The amplitude of this pattern is adjusted each iteration using a feedback controller with visual camera measurements. In the next part of this process, the plate is reshaped by applying the punch at a discrete location on the plate at each iteration. A geometric optimization problem is formulated to model how the plate is reshaped given a small displacement at a single point. Another optimization problem is then formulated to determine the point on the plate to displacement and the amount of this displacement. Then, visual camera-based feedback control is used to incrementally adjust the punch displacement to account for plate springback. The reshaping process continues until the plate achieves an acceptable morphology. In the third process a blue light scanner is used to measure the plate morphology after each iteration in a robotic incremental sheet forming operation. A learning controller is used to update the tool path based on the morphology measurements.

Keywords: Manufacturing Systems, Mechatronic Systems, Control Applications
Abstract: Digital Glass Forming (DGF) is a heat assisted manufacturing process that uses a laser to create a workable zone that is subsequently shaped. In additive DGF processes, the glass is continuously fed via a filament or fiber into the work zone on a moving substrate and the work zone material is added to the substrate or previously deposited material. The DGF process is a thermomechanical process that requires the work zone temperature to be high enough such that the glass is formable but low enough that the glass does not vaporize. Further, since the DGF process is a forming process, the work zone can deform and move during deposition and, thus, care must be taken such that material is in the correct location after processing. In this talk we will discuss the use of learning control to regulate the temperature and material placement in DGF. In the first study, learning control is combined with a feedback controller when fabricating walls. The learning controller adjusts the laser power quickly at the corners to prevent evaporation while the feedback controller accounts for variations from print to print. The morphology of the freestanding structure with a joint is regulated via learning control. The deposition path and the distance of the work zone from the laser focal point are adjusted to achieve good placement of the strut and joint, respectively. In third study, the learning control is applied to the reflow process to regulate the height of wall depositions.

Keywords: Manufacturing Systems, Modeling and Validation
Abstract: Generative AI and foundation models (FMs) have demonstrated strong capabilities in knowledge representation, reasoning, and human–machine interaction. However, general-purpose models are not inherently equipped with the process-specific knowledge, historical information, and engineering context required for reliable decision support in manufacturing systems. Effective integration therefore requires methods for adapting these models to manufacturing domains and connecting them with relevant data and computational resources. This work investigates a framework for developing a manufacturing-oriented GenAI assistant by combining foundation-model adaptation, including fine-tuning, with supporting information-retrieval and computational workflows. Extrusion-based additive manufacturing is considered as an initial case study for developing and evaluating the proposed approach. This study examines how manufacturing-specific adaptation can improve the relevance, contextual understanding, and reliability of model responses while extending their capabilities beyond conventional knowledge retrieval. By connecting generative AI with historical manufacturing information, engineering knowledge, and computational resources, the proposed approach can provide a more informed interface for manufacturing decision support. Such capabilities may further support virtual commissioning and digital manufacturing environments by improving interaction between users, data, models, and virtual representations of physical systems.

Keywords: Manufacturing Systems, Modeling and Validation, Control Applications
Abstract: Aerosol Jet Printing (AJP) is a promising additive manufacturing process for high-resolution printing of functional materials. However, maintaining consistent print quality remains challenging because the process is highly sensitive to operating conditions such as aerosol and sheath gas flow rates. While previous studies have primarily focused on steady-state relationships between process parameters and printed features, the lack of dynamic process models has limited the development of model-based closed-loop control.

This work presents a physics-informed dynamic modeling framework for Aerosol Jet Printing based on transient computational fluid dynamics and particle tracing simulations. A time-dependent COMSOL model is developed to investigate particle transport under dynamically varying process conditions. Particle impact characteristics, including radial impact velocity and impact angle, are analyzed as physically interpretable state variables linking process inputs to printed linewidth and overspray. Transient input excitations are further used to characterize the dynamic response of these variables and support system identification.

Using the simulated transient data, a reduced-order state-space representation is identified to describe the input-to-state and state-to-output dynamics. The results demonstrate that the selected particle states capture the input-driven transport dynamics, while also revealing limitations in predicting deposition-based output metrics. These observations provide insight into the selection of additional physical states and model structures required for improved output prediction. The proposed framework provides a pathway from high-fidelity multiphysics simulation to control-oriented dynamic models for future model-based closed-loop control of AJP.

Keywords: Mechatronic Systems, Manufacturing Systems, Sensors and Actuators
Abstract: While additive manufacturing has become an enabling technology, existing techniques are not capable of directly 3D printing high current electromagnetic motors. In this work, a novel 3D-printable, multi-layer, wave-winding electromagnetic actuator topology is created for generating high efficiency rotary electric motors. This is achieved through the development of a new additive manufacturing technique that incorporates conductive silver nanoparticle inks, thermally conductive but electrically insulating polymers, and surface mount electrical components. Excluding the permanent magnets, this work is the first to demonstrate the use of 3D printing for manufacturing all of the motor elements. This includes the stator, housings, bearings, and sensing circuits for an axial flux motor, cylindrical radial flux motor, and a linear motor. Results show that the axial flux motor can reach a peak stall torque of 11.75 N·mm/A and a maximum efficiency of 40.4%. These results represent improvements of approximately 7.8× in torque and 5.3× in efficiency when compared with previously reported 3D-printed motor designs. Thermal time constant measurements and dynamic response measurements show that this motor can operate up to 180ºC and a bandwidth up to 5.2 Hz. Using these unique 3D printed actuators, realistic applications are demonstrated including a fan, water pump, paddle-wheel boat, multi-legged walking robot, and waving arm. With the material and design innovations discussed in this work, the creation of fully 3D-printed electromechanical devices, machines, and robots become possible.

Keywords: Mechatronic Systems, Path Planning and Motion Control, Manufacturing Systems
Abstract: Continuous-fiber-reinforced polymer (CFRP) composites offer exceptional strength-to-weight performance for structural applications in aerospace, automotive, and architectural engineering. Additive manufacturing (AM) with robotic systems enables fabrication of complex CFRP geometries with customized fiber placement, yet achieving high mechanical performance and geometric fidelity simultaneously remains a fundamental challenge. This talk presents integrated advances in robotic CFRP-AM spanning path planning, structural optimization, and real-time process control.

On the path planning front, a gradient-based B-spline finite element approach is introduced for continuous fiber path optimization, employing a semi-analytical Fiber-in-Element Probability function to efficiently compute structural responses and gradients with respect to fiber path control points. To address variable-width deposition in freeform geometries, a streamline-based fiber path generation method combined with Voronoi-diagram-based matrix infill planning dynamically adapts deposition width to local geometric constraints, improving fill quality and interfacial bonding.

For three-dimensional shell structures in multi-axis robotic AM, an Evolving Stress-Guided (ESG) framework is proposed, iteratively updating the structural stress field as fibers are deposited layer by layer. Coupled with bijective quasi-conformal surface mapping and a one-stroke toolpath algorithm, the framework produces manufacturable multi-axis trajectories with significantly enhanced structural stiffness. The conformal mapping strategy further enables fiber path design in a 2D domain projected onto complex curved surfaces for robotic execution.

To close the loop between planning and manufacturing, a contact-force-based closed-loop control system provides real-time process monitoring and correction. A six-DOF load cell integrated at the robotic end-effector delivers high-frequency force feedback, regulating fiber alignment via tangential and side force signals and interlayer bonding via vertical force feedback. A PD controller dynamically adjusts layer height during concave surface deposition, effectively eliminating fiber bridging defects.

Keywords: Cyber physical systems, Intelligent Autonomous Vehicles, Machine Learning in modeling, estimation, and control
Abstract: The reliability of autonomous vehicles (AVs) has become increasingly important. In this paper, we propose a reinforcement learning adversarial testing framework for analyzing vulnerabilities in self-driving policies. We model adversarial agents that are capable of causing collision events in a target vehicle controlled by a self driving algorithm Follow-the-Gap (FTG). We introduce Gap Lock, a reward to encourage actions that lead to wall impacts. Simulation results show attackers reliably degrade FTG performance and cause crashes in scenarios where FTG remains safe under nominal conditions. We then demonstrate model transfer from simulation to real-world deployment on a physical platform.

Keywords: Estimation, Machine Learning in modeling, estimation, and control, Control Applications
Abstract: This paper proposes a Koopman-based state-of-charge (SOC) observer for lithium-ion batteries designed in a lifted linear space identified via extended dynamic mode decomposition with control (EDMDc). Accurate SOC estimation still remains challenging, since battery electrochemical dynamics are highly nonlinear and the SOC cannot be directly measured. To capture the intrinsic nonlinear characteristics of battery dynamics, we construct a physics-informed dictionary of observable functions defined over the battery state variables. Based on the resulting Koopman linear representation, battery SOC observers are developed via linear matrix inequalities (LMIs), ensuring the stability of the estimation error dynamics in the lifted space. Finally, the proposed framework is validated using multiple battery datasets under various dynamic loading conditions. The proposed observer achieves root mean square errors (RMSE) between 0.0214 and 0.0248 across different driving profiles, demonstrating significantly improved SOC estimation performance compared with an extended Kalman filter (EKF)-based approach (RMSE: 0.0401–0.0505), which has been widely used in the battery community.

Keywords: Estimation, Machine Learning in modeling, estimation, and control, Sensors and Actuators
Abstract: This work focuses on the estimation of state and parameters for a soil moisture monitoring system comprising an above-ground transmit coil inductively coupled to a buried passive sensor. The system is modeled as a coupled resistor-inductor-capacitor (RLC) oscillator. A dual extended Kalman filter (DEKF) is developed to simultaneously estimate the states and unknown parameters to infer soil-moisture level. Specifically, the separation distance between the buried sensor and a concentrically aligned transmit coil is estimated through the mutual inductance between the two circuits, while the soil-moisture level is determined by estimating the buried sensor's self-capacitance. It is shown that extending the maximum correntropy criterion (MCC) to the DEKF improves estimation performance in the presence of unmodeled dynamics arising from system implementation complexities. The approach is validated through hardware-in-the-loop experiments. In fact, the MCC-DEKF, on average, resulted in 3.33-times faster parameter-estimate convergence, as well as reduced average estimate errors of the sensor separation distance by 3.05~cm and the soil moisture by 1.78 percentage points, compared to a standard DEKF. These results can be used to develop an integrated autonomous irrigation system for precision agriculture to monitor and regulate soil moisture levels.

Keywords: Estimation, Machine Learning in modeling, estimation, and control, Uncertain Systems and Robust Control
Abstract: Abstract Time-varying delays are a common challenge in many control systems and can significantly degrade a system's performance and stability. This challenge is compounded in nonlinear systems that contain uncertain dynamics and experience unknown time-varying output delays, where the true system state measurements are delayed. To address these issues, this paper develops a dual observer framework in which one observer estimates the uncertain nonlinear dynamics, while a second observer reconstructs the true, delay-free system states. The proposed approach leverages a neural-network (NN)-based observer framework to approximate the unknown dynamics and to compensate for the uncertain output delays by predicting the true states. A Lyapunov-like stability analysis is used to show that the proposed observers yield semi-global exponential convergence of the estimation errors to a uniformly ultimately bounded set around 0. Numerical simulations are provided to demonstrate the effectiveness of the proposed observers in accurately estimating the true system states, despite the presence of model uncertainties and output delays.

Keywords: Machine Learning in modeling, estimation, and control
Abstract: Markerless motion capture systems such as OpenCap have increased the accessibility of human movement analysis by estimating joint kinematics from smartphone videos. However, residual errors in these estimates can propagate through downstream musculoskeletal analyses, reducing the reliability of derived quantities such as joint moments and contact forces. This study investigated data-driven restoration of lower-extremity joint angles estimated by OpenCap, using synchronized marker-based motion capture as the reference. We compared three sequence modeling architectures with distinct inductive biases: (1) a CNN–BiLSTM autoencoder operating on full-trial sequences (285K parameters), (2) a Transformer encoder with center-segment residual correction applied to sine/cosine windows (2.3M parameters), and (3) a two-stage VAE-Bridge framework that leverages a learned clean-motion prior for latent-space correction (3.7M parameters). The benchmark dataset comprises 594 curated trials (1,204,082 frames) from 20 participants performing 10 lower-extremity exercises. Performance was evaluated using leave-one-exercise-out (LOEO) and leave-one-subject-out (LOSO) cross-validation, with metrics including circular mean absolute error (MAE). Under LOEO evaluation, the Transformer achieved the lowest overall MAE of 3.77◦ (26.3% reduction from the 5.12◦ baseline of OpenCap estimation). Under LOSO evaluation, the VAE-Bridge achieved the lowest overall MAE of 3.85◦ (24.7% reduction from the baseline).

Keywords: Machine Learning in modeling, estimation, and control, Control Design, Adaptive and Learning Systems
Abstract: Reinforcement learning (RL) can improve nonlinear control for safety-critical systems, but it may produce unsafe actions when trained without an explicit safety structure. This paper presents a CBF-guided safe RL framework for the QUBE-Servo 2 rotary inverted pendulum as a hardware benchmark for safety-critical control. A second-order control barrier function is derived from the arm-angle constraint and used to compare unconstrained RL, shielding-only RL, CBF-guided training with shielding, and CBF-guided training without the deployment filter. Simulation and hardware results show that CBF-guided training produces smoother stabilization and reduces dependence on runtime correction. An event-triggered runtime assurance mode further reduces CBF-QP activation while preserving safe behavior, supporting the use of training-time safety shaping for learning-enabled safety-critical systems.

Keywords: Automotive Systems, Estimation
Abstract: This paper proposes a voltage‑aware electrochemical impedance spectroscopy (EIS) approach for early detection of soft shorts in lithium‑ion battery cells. The method exploits the sensitivity of low‑frequency impedance characteristics and voltage‑dependent changes in the impedance spectrum to leakage currents caused by high‑resistance soft shorts. By analyzing impedance features as a function of cell voltage, the method reduces the influence of state‑of‑charge variation and normal aging effects without relying on explicit electrochemical models or dynamic load profiles. All results are obtained from controlled single‑cell experiments under different voltage levels and externally induced soft‑short conditions. The proposed method enables model‑light, charge‑agnostic fault screening technique at the cell level, offering practical insights for early safety assessment and future extension to battery pack applications.

Keywords: Automotive Systems, Estimation
Abstract: Lithium‑metal batteries (LMBs) exhibit pressure‑dependent behavior that reflects both electrochemical activity and mechanical degradation, making pressure a valuable indicator of cell health. To avoid the cost and packaging penalties of physical sensors, this paper presents a data‑driven framework for sensorless pressure estimation. The method incorporates voltage, current, temperature, SOC, and normalized solid‑phase lithium concentration to model pressure evolution. A normalized pressure formulation is introduced to remove dependence on initial mechanical setup. A recurrent neural network (RNN) is then developed to capture nonlinear temporal dynamics. Characterization and validation results demonstrate accurate and robust pressure estimation across diverse operating conditions.

Keywords: Automotive Systems, Modeling and Validation, Transportation Systems
Abstract: New thermal runaway (TR) propagation prevention requirements have accelerated the development of advanced battery pack insulation materials. In this work, we develop a method to extract the lumped thermal conductivity properties of insulation materials while accounting for multi-modal heat transfer, including inter-cell conduction, inter-tab conduction, and convection to the surroundings and the coolant. The identification of the effective thermal conductivity of the various thermal pathways under realistic (pack) pressure relies on a highly instrumented two-cell mini-module with a heater attached on one of the cells. The set-up provides valuable data on the TR under overheating, and its simplicity supports the unique identification of conductivity parameters that can subsequently feed more complex and high-fidelity pack simulations as an efficient workflow for investigating propagation dynamics. As an example of the workflow, we show how the identified insulation parameters from one experiment are used to find the peak temperature of the adjacent cell when the thickness of insulation material doubles or in the presence of electric tabbing and coolant. More importantly, we show that the identifiability of the insulation material is weakened when electric tabbing and cooling are involved, supporting the use of the simple two-cell fixture for identification purposes.

Keywords: Machine Learning in modeling, estimation, and control, Automotive Systems, Estimation
Abstract: Accurate state-of-health estimation of lithium-ion batteries is critical for the safe operation of electric vehicles, yet deploying high-fidelity prognostic models on a vehicle’s onboard battery management system is often challenging due to limited computational resources. To address this challenge, we propose a novel, lightweight physics-informed reservoir computing (PIRC) framework that integrates physical degradation priors directly into a closed-form ridge regression readout. PIRC employs an adaptive switching mechanism that autonomously selects the optimal feature representation, mitigating over-parameterization instability while strictly minimizing the number of necessary trainable parameters. We demonstrate PIRC’s effectiveness on the NASA PCoE battery dataset and benchmark chaotic systems, achieving lower mean error with fewer trainable readout weights, resulting in lower computational overhead than standard physics-informed neural networks and existing hybrid reservoir models.

Keywords: Modeling and Validation, Control Applications
Abstract: This paper presents a Data-Enabled Predictive Control (DeePC) framework for immersive battery thermal management systems (IBTMS) using high-fidelity computational fluid dynamics (CFD) simulation data. The proposed method avoids explicit model identification by directly constructing a predictive controller from historical input–output trajectories. To evaluate the effectiveness of the approach, CFD models of both the IBTMS and a conventional cold-plate battery thermal management system (CPBTMS) are developed in ANSYS Fluent under both hot and cold operating conditions. The DeePC controller regulates the battery temperatures by adjusting the inlet coolant velocity and inlet temperature. Simulation results demonstrate that the proposed framework can achieve accurate temperature tracking while maintaining smooth control inputs. Compared with the CPBTMS, the IBTMS shows a significantly faster thermal response and better temperature uniformity. In addition, the IBTMS requires a shorter data collection period and results in lower online computational cost, indicating its strong potential for real-time battery thermal management.

Keywords: Human-Machine and Human-Robot Systems, Machine Learning in modeling, estimation, and control, Intelligent Autonomous Vehicles
Abstract: Reliable take-over in SAE Level 2 automated driving depends not only on system status but also on how drivers perceive surrounding risks. However, existing Human-Machine Interaction (HMI) designs still lack explicit and interpretable representations of driver attention, limiting the understanding of how drivers allocate attention during maneuvers. To address this issue, this study develops an adaptive grid-based representation of driver gaze attention for take-over time analysis. A driving simulator experiment with eye tracking was conducted with 12 participants under dynamic traffic conditions. Within a grid-based traffic framework, gaze behavior is mapped into attention weights associated with surrounding traffic, enabling visualization of driver attention distribution and providing an interpretable link between driver perception and traffic interactions. In particular, the proposed representation captures both spatial and temporal characteristics of driver attention, while the adaptive weighting strategy further emphasizes high-attention regions and reduces the influence of negligible grids. These findings provide a more transparent and human-centered perspective for analyzing take-over time in automated driving systems.

Keywords: Intelligent Autonomous Vehicles, Path Planning and Motion Control, Automotive Systems
Abstract: This study proposes an automated driving system for navigating narrow L-shaped roads with intersection angles other than 90° using iterative reversing maneuver using LiDAR based on localization and path-following control. This system is based on a combination of path-following control and the pre-boundary stopping control method. Driving experiments using a scale model vehicle showed that the vehicle successfully passed all tested road conditions without colliding with the walls, demonstrating the applicability of the proposed system to real vehicles.

Keywords: Machine Learning in modeling, estimation, and control
Abstract: Accurate and efficient prediction of physical fields with limited data has become a fundamental yet challenging problem in complex structural systems. Intricate geometries and heterogeneous materials, particularly in lightweight structural designs involving geometric discontinuities, often induce strong nonlinear and localized responses, posing significant challenges for accurate prediction. Thus, we propose StructGTO, an efficient graph transformer operator for physical field prediction in complex structures with geometric discontinuities. Specifically, we propose a hybrid embedding scheme that jointly encodes graph topology, spatial coordinates, and implicit geometric information into a unified representation over nodes and edges for downstream graph transformer prediction. Experimental results show that the proposed method achieves high-fidelity physical field prediction compared to existing methods.

Keywords: Machine Learning in modeling, estimation, and control, Estimation, Modeling and Validation
Abstract: Gear systems are very common in the mechanical world, making the proper conduction of prognosis and diagnosis an important task. Data-driven analysis has become dominant in this field as vibration data is easy to collect and is rich in information about its health condition and operating speed. Moreover, Machine Learning (ML) architectures are often leveraged to conduct this analysis due to its capabilities in pattern recognition and classification. However, data from real operating systems provides very little information on faulty conditions, and data from experimental testbeds can only provide a very limited amount of fault categories and severities to train an ML architecture. While both have their advantages, a gearbox can develop a nearly infinite amount of fault type and severity combinations, making this lack of label diversity a big obstacle. To address this challenge, this paper presents a modified convolutional neural network that leverages well documented Bayesian Inference techniques to implement probabilistic weights and biases into the convolution process. This Bayesian Neural Network (BNN) will aim to observe faults that were not present during its training process and estimate their severity based on a reduced fault library used for training. The paper leverages the use of a 26 DOF lumped parameter model to generate 30 chipped tooth severities, some of which will be used to train the model and the rest to simulate unseen faults.

Keywords: Modelling, Identification and Signal Processing, Machine Learning in modeling, estimation, and control, Manufacturing Systems
Abstract: Maintaining product quality in continuous steel casting requires reliable prevention of slag carry-over into downstream components. This paper proposes a vibration-based approach for the detection of slag-related events based on acceleration signals measured on the shroud manipulator. From these signals, time-series features are extracted and used as inputs to a recurrent neural network (RNN) within an anomaly detection framework. The RNN is trained to model nominal system behavior, and deviations from its predictions are used as indicators of potential slag intrusion. To improve sensitivity to relevant events, a dedicated evaluation function is introduced to highlight significant changes in the signal dynamics. The approach is evaluated on a limited set of industrial measurements and demonstrates its potential as a practical and low-cost solution for real-time monitoring applications.

Keywords: Robotics, Adaptive and Learning Systems, Machine Learning in modeling, estimation, and control
Abstract: Visual inspection is widely used in industrial quality control to detect defects, verify structures, and ensure product consistency. In many existing robotic inspection pipelines, reliable observation is achieved through fixed fixtures, predefined robot motions, and controlled viewpoints, so that the region of interest can be consistently exposed to the sensor. While effective in structured production environments, such workflows are less suitable for high-mix low-volume (HMLV) settings, where product geometries, inspection targets, and initial object poses may vary across instances. In these cases, a fixed observation setup may fail to expose the task-relevant region, limiting inspection quality even when the downstream vision model is strong. To address this limitation, we study inspection-oriented pose preparation before inspection. We formulate the problem as a task-driven action selection problem, where the system predicts a discrete object reorientation action to improve the visibility of the target region under the current observation and task requirement. We propose a hierarchical framework in which a vision-language model performs high-level inspection-aware decision making from two top-view RGB observations and the inspection target, while an ACT-based low-level policy executes the selected primitive on the robot. We evaluate the proposed framework on a robotic setup with predefined reorientation primitives. Compared with conventional exhaustive inspection routines and rule-based baselines, the proposed approach provides a more flexible target-conditioned decision process that reduces unnecessary motions and better adapts to varying initial object poses and inspection requirements.

Keywords: Control Applications, Control Design, Robotics
Abstract: Weeds remain a major challenge in agriculture, particularly with the rise of herbicide-resistant species and the growing demand for organic production systems. As a result, mechanical weeding has become an important alternative to chemical control. In this study, a site-specific mechanical weeding (SSMW) robot was developed, featuring autonomous, self-adjustable, individually controlled sweeps for precision tillage and targeted weed removal. A strain-based load cell was integrated into the individual shank control to measure sweep-soil interaction forces during operation. The performed field experiments showed uncertainty and nonlinearity in the strain measurements, which resulted in inaccuracies in the feedback control of the SSMW system. To resolve this, an adaptive baseline estimation approach was implemented to correct the strain-reading-based feedback by referencing the actuator position using a laser distance sensor. Add one comment on the successful realization of the sensor fusion and the accuracy of the control. The proposed baseline correction method reduced the baseline error by 97.9%.

Keywords: Control Applications, Optimal Control, Soft Robotics
Abstract: We present an iterative algorithm for shape control of deformable structures based on multi-modal density steering methods. In our approach, the dynamics of a deformable structure are described by a mass-spring-damper (MSD) network, and its shape is represented by a multi-modal density model, and in particular, a Gaussian Mixture Model (GMM). This modeling setup allows us to pose the shape control problem as a GMM (density) steering problem, eliminating the point-to-point correspondence required by conventional configuration control methods. The proposed density steering algorithm for shape control must be applied iteratively because the dynamics of the MSD network are nonlinear. The convergence of the proposed algorithm is demonstrated through extensive numerical simulations (Monte Carlo analysis). Benchmarking against a basic Model Predictive Control (MPC) scheme shows that our proposed shape control approach achieves significantly faster solve times while introducing only a small point-wise position error, offering a scalable alternative for high-dimensional deformable systems.

Keywords: Control Design, Robotics, Mechatronic Systems
Abstract: Robotic welding of multi-seam parts requires accurate seam following, stable tool orientation, and constant tangential speed, but commercial robot interfaces typically expose only pose- or velocity-streaming modes, creating a practical tradeoff between geometric fidelity and speed regulation. This paper presents a novel coupling-injected constant-speed path-following (CSPF) controller and evaluates its performance in multi-seam robotic welding. Relative to the baseline CSPF formulation, the proposed controller explicitly enforces tangential phase convergence while preserving the constant-speed path-following objective. The analysis further shows that the curvature-dependent coupling nonlinearity, often canceled in pursuit of linearized dynamics, can instead be exploited as a stabilizing design feature. The controller is implemented within a computer-aided design (CAD)-to-robot welding workflow comprising workpiece registration, path planning, and multi-seam execution. Dry-runs, benchmark comparisons, and welding experiments show improved path following accuracy, tangential-error regulation, and seam-end consistency without degrading spatial path following performance.

Keywords: Robotics, Path Planning and Motion Control, Control Applications
Abstract: The problem of realizing hopping locomotion of a four-link planar robot, hopping on rigid ground, is addressed. The complete dynamic model that captures the flight, impact, and stance phases of hopping is presented, and transformed to normal form. Partial feedback linearization is used to regulate the position of the center-of-mass with respect to the toe to desired setpoints. Simulations show that appropriate choice of setpoints, and dynamic update of the setpoints at each impact can be used to realize locomotion with a desired apex height and a desired step size.

Keywords: Robotics, Soft Robotics, Control Applications
Abstract: This paper presents a sensorless soft-catching control for a mobile manipulator used for ball catching in robot lacrosse. We solve the ball-bouncing-out-of-net problem by studying pre-contact and post-contact dynamics when the ball interacts with the net. We formulate the control problem and use Lyapunov stability analysis to establish conditions to tune the feedback control gains. Real robot experiments demonstrate an increased ball catching rate without extra sensors.

Keywords: Robotics, Estimation, Control Applications
Abstract: Unknown payloads can strongly affect compliant robotic manipulation, especially when the payload center of mass is not aligned with the tool center point. In this case, the payload generates an offset wrench at the robot wrist. During motion, this wrench is not only related to payload weight, but also to payload inertia. If it is not modeled, the compliant controller can interpret it as an external interaction wrench, which causes unintended compliant motion, larger tracking error, and reduced transport accuracy. This paper presents a wrench-aware admittance control framework for unknown-payload pick-and-place using a UR5e robot. The method uses force-torque measurements in two different roles. First, a three-axis translational excitation term is used to reduce payload-induced force effects during transport without making the robot excessively stiff. Second, after grasping, the controller first estimates payload mass for transport compensation and then estimates the payload center-of-mass offset relative to the tool center point using wrist force-torque measurements collected during the subsequent translational motion. This helps improve object placement and stacking behavior. Experimental results show improved transport and placement performance compared with uncorrected placement while preserving compliant motion.

Keywords: Agricultural Systems, Optimal Control, Machine Learning in modeling, estimation, and control
Abstract: Greenhouse supplemental lighting is essential for meeting crop Daily Light Integral (DLI) requirement. However, conventional always-on and rule-based strategies are inherently energy-inefficient since they ignore future sunlight availability. This paper reports the results of deploying various supplemental lighting control strategies, namely, always-on, always-off, heuristic, optimal prediction-based, and two peak-shift variants, simultaneously in a single greenhouse at the Hampton Roads AREC of Virginia Tech, to ensure that every treatment experiences identical conditions. The prediction-based and peak-shift strategies optimize LED schedules using a Markov chain-based solar forecast. Experimental results over two replicates show that the prediction-based strategy meets the desired DLI target at 65% cost reduction relative to the always-on baseline while achieving the highest dry-mass yield among the proposed adaptive strategies. The heuristic strategy incurs only a 6% cost premium over prediction-based control, offering a competitive forecast-free alternative. The peak-shift variants achieve the lowest active-strategy costs by shifting LED demand into off-peak tariff windows. A yield-per-dollar efficiency metric is finally introduced to jointly capture energy cost and crop output showing that the prediction-based strategy ranks highest at 18.976 g/, more than double the always-on value of 9.391 g/. These findings can provide greenhouse operators with an evidence-based framework for selecting a lighting control approach that balances energy cost, crop quality, and system complexity.

Keywords: Control of Smart Buildings and Microgrids, Modelling and Control of Environmental Systems, Optimal Control
Abstract: Maintaining indoor particulate matter (PM) concentrations below recommended levels is crucial for indoor air quality (IAQ) in buildings. This paper proposes control strategies to adhere to these recommended thresholds with minimal energy consumption from ventilation systems. For single-zone buildings, the proposed steady-state optimal controller that uses both indoor and outdoor air for IAQ management achieves a 10% energy reduction compared to a baseline controller under realistic PM variations, originating from either indoor generation (e.g., cooking) or outdoor infiltration (e.g., urban vehicle traffic). The study further expands on control strategies for multi-zone buildings and demonstrates an achievable energy reduction of 15-33% by utilizing airflow from other “cleaner” zones to meet the IAQ target of a specific zone with high PM generation rate. This approach relies on airflows to/from different zones through a central duct, i.e., does not require direct air mixing between the zones. The steady-state optimal solution is observed to attain results comparable to a model predictive control strategy, but at much lower computational complexity.

Keywords: Linear Control Systems, Optimal Control, Control Design
Abstract: Model predictive control (MPC) is a promising strategy for lithium-ion battery management, but practical implementation is limited by computational complexity and nonlinearities such as voltage hysteresis. Dual-mode MPC (DMPC) ensures stability via terminal invariant sets, but only within a restricted controllable region. This paper introduces a pre-conditioning stage that steers the system toward this region, improving the applicability of DMPC. An electro-thermal model with hysteresis captured through affine terms is used. Terminal invariant and N-step controllable sets define switching conditions. The proposed method improves charging performance and reduces computation while preserving constraint satisfaction and stability guarantees within the DMPC region.

Keywords: Optimal Control, Automotive Systems, Control Applications
Abstract: This paper aims to advance state-of-the-art energy management strategies by developing energy-optimal and computationally tractable control methods for electric motors that can be independently controlled. The proposed method jointly optimizes driving torque and torque allocation between the front and real axle motors while ensuring satisfaction of system constraints. To enable real-time applicability, we develop an analytical expression of power consumption that takes into account the motor efficiency and formulate the convexified optimal control problem. Additionally, we study a tractable learning-based control strategy in the presence of parametric uncertainty and actuation errors. The proposed method is implemented and validated in scalable high-fidelity co-simulation experiments with urban traffic disturbances. The real-time applicability of the method is validated in electric control unit hardware-in-the-loop (ECU-HIL) experiments. The results shed lights on the real-world impact of the proposed method on energy efficiency. Simulation results show up to 7.79% energy efficiency improvement and reduction in computing time. Real-time applicability is also validated in ECU-HIL experiments: the computing time of the proposed method is substantially smaller than the simulation sampling period.

Keywords: Optimal Control, Nonlinear Control Systems
Abstract: Optimal control plays a critical role in nonlinear dynamical system design, yet traditional methods can present practical challenges: indirect methods often require good initial guesses for the costates, while direct collocation methods can suffer from conditioning issues and inefficient mesh refinement. In this work, a ph-adaptive shared-node cover scheme is paired with Coupled Radial Basis Functions (CRBFs) to provide structured discretization refinement for solving optimal control problems. The methodology consists of employing an indirect formulation to derive the necessary conditions, local CRBF-collocation over segments of the problem domain, shared nodes between segments to enforce continuity, and hierarchical ph-refinement driven by state error estimate. The framework enables controlled discretization growth while maintaining solution accuracy. Numerical simulations of the nonlinear Duffing oscillator and the inverted pendulum on a cart demonstrate that the proposed approach achieves prescribed error-estimate tolerances while closely matching the ode45-propagated dynamics. This work presents the first adaptation of a p-then-h-inspired refinement strategy within a CRBF-based cover scheme, prioritizing shared-node refinement before nodal and element refinement and providing a foundation for future computational cost analysis and constrained optimal control applications.

Keywords: Optimal Control, Robotics, Control Design
Abstract: Sampling-based model predictive control (MPC) methods such as the Model Predictive Path Integral (MPPI) controller use isotropic Gaussian noise to explore the control space, ignoring the local cost landscape geometry. We integrate the covariance-optimal (CoVO) sampling strategy into a GPU-accelerated MPC framework operating on spline-parameterized control trajectories within MuJoCo XLA (MJX) environments. Our method, CoVO-MPPI, uses weighted sample covariance as a proxy for inverse local curvature and shapes the sampling covariance so that exploration concentrates along high-sensitivity directions while preserving the total sampling volume. On five MuJoCo benchmark tasks (pendulum, cart-pole, crane, particle, walker), CoVO-MPPI reduces cumulative cost by 2-6% on tasks with anisotropic cost landscapes relative to single-iteration MPPI, at the same per-step wall-clock cost and with no additional samples or simulation rollouts, while avoiding the 30-50 s autodiff-Hessian evaluation required by the original CoVO-MPC formulation; however, it can degrade or fail to improve performance on tasks with lower anisotropy or insufficient N/d ratio. We validate the sample-based estimator against the exact autodiff Hessian and identify cost landscape anisotropy (measured by the Hessian condition number) and the sample-to-dimension ratio N/d as the two factors governing when adaptive covariance helps. The implementation is open-source and built on top of the Hydrax and JAX ecosystem.

Keywords: Control Applications, Optimal Control, Transportation Systems
Abstract: This paper proposes a learning-based multistage model predictive control (LB-msMPC) framework for macroscopic highway traffic control. Traffic dynamics are modeled by a nominal METANET system with variable speed limit (VSL) actuation and an additive state- and input-dependent mismatch term. The mismatch is learned using local Gaussian Process Regression (GPR) models trained on data pooled across cells and evaluated using local cell-neighborhood regressors. The resulting GPR posterior defines a state-dependent stochastic disturbance embedded in a scenario-tree msMPC. Compared with worst-case robust MPC and no control, the proposed approach reduces conservativeness while improving mobility, achieving lower total travel time and higher throughput.

Keywords: Machine Learning in modeling, estimation, and control, Transportation Systems, Large Scale Complex Systems
Abstract: Pavement deterioration is a dynamics process, involving the continuous, progressive decline of structural and functional integrity over time. In a road network, accurate multi-year pavement deterioration forecasts are essential for optimizing network-level maintenance programming and ensuring cost-effective budget planning. Using Pavement Management System data, this paper compares four strategies across four distress targets: a persistence baseline, parametric age curves, direct-horizon machine-learning (ML) models, and a recursive rolling ML system. Direct ML provides the highest accuracy at 3-10 year horizons, while parametric curves underperform on the full sample but are adequate for the central 80% of sections. The top 10% of sections account for 56-79% of total squared error, and recursive ML suffers from error compounding at long horizons. The findings provide guidance on when to favor simple baselines, calibrated parametric curves, or data-driven ML under real-world data quality constraints.

Keywords: Path Planning and Motion Control, Optimal Control, Transportation Systems
Abstract: This paper presents a simultaneous target sequencing and trajectory optimization framework for energy-optimal drone delivery. By leveraging Pontryagin’s Minimum Principle (PMP) and a continuous Gaussian reward field, the approach enables the delivery drone to prioritize high-reward regions while minimizing overall energy consumption. Optimal rotor thrusts are computed in closed form with backward costate integration, enabling real-time performance. Simulations show reduced energy usage compared to sequential optimization. The results demonstrate the potential of PMP-based simultaneous target sequencing and trajectory optimization for rapid, energy-aware drone delivery systems.

Keywords: Power and Energy Systems, Control Applications, Transportation Systems
Abstract: Hybrid power trains (HPT) run on multiple energy sources, often involving energy storage systems/batteries (ESS). As a result, the risk of battery degradation and the reliability of energy storage elements pose a major challenge in designing an energy-efficient hybrid power train. Incorporating the battery degradation model directly into the underlying optimization problem is challenging on multiple fronts: 1) Any reasonable degradation model will, due to its complexity, result in a complicated optimization problem that is impractical for real-time implementation. 2) The models contain a lot of time-varying parameters that can only be determined through destructive experimental procedures. As a result, it is essential to devise heuristics that reasonably capture the degradation per usage of the batteries. One such heuristic considered in this paper is the absolute power extracted from the battery. A distributed model predictive strategy is then developed to coordinate the power split to maximize efficiency while mitigating the failure risk due to battery degradation. The designed energy management (EM) strategy is demonstrated through a realistic simulation of a hybrid power train: hybrid road vehicles (for example, a hybrid electric vehicle (HEV)). The results show the effectiveness of the energy management strategy in managing battery degradation.

Keywords: Automotive Systems, Path Planning and Motion Control, Optimal Control
Abstract: High-speed heavy-duty semi-trailers present significant control challenges due to their pronounced nonlinear dynamics, parameter uncertainties, and the difficulty of accurately acquiring real-time kinematic states. To ensure trajectory tracking precision and lateral stability, this paper proposes an advanced control framework integrating Nonlinear Model Predictive Control (NMPC) with an Extended State Observer (ESO). A nonlinear vehicle dynamics model is established as the predictive foundation, while the ESO is formulated to encapsulate both internal parameter perturbations and external trailer-induced coupling forces into a unified lumped disturbance for real-time feedforward compensation. Furthermore, a multi-point look-ahead strategy is introduced to capture high-order dynamic behaviors, complemented by an NMPC architecture that inherently incorporates physical constraints to safeguard vehicle stability. Extensive MATLAB/Simulink and TruckSim co-simulations under varying payloads (0, 20, and 40 tons) during DLC and cornering maneuvers demonstrate that the NMPC-ESO significantly outperforms LQR, MPC, and standard NMPC. The proposed strategy achieves the minimum lateral RMSE and optimal driving stability across all loading conditions, validating its superior robustness.

Keywords: Optimal Control, Automotive Systems, Modeling and Validation
Abstract: Even with the widespread adoption of electric vehicles, thermal management of the powertrain under extreme conditions remains a critical challenge. As such, powertrain controllers limit motor torque to protect components from high temperatures, a process commonly referred to as thermal derating. This paper proposes a novel thermal derating strategy for electric powertrains based on model predictive control (MPC). Unlike conventional approaches in the literature that directly modify the requested motor torque, the proposed method dynamically adjusts the upper bound of the commanded torque using a data-driven linear time-invariant (LTI) model of the motor thermal dynamics. The proposed approach substantially reduces computational burden by operating at a control loop two orders of magnitude slower than that of the torque controller. Further, it enhances safety by predictive management of temperature constraints over an extended horizon while maximizing propulsion capability.

Keywords: Cyber physical systems, Control Design, Linear Control Systems
Abstract: Real-time hybrid simulation (RTHS) is an efficient and cost-effective method for evaluating real-world structural behavior of cyber-physical systems. This research expands on the principles of RTHS to lay the groundwork for the application of this experimental methodology for rapid control prototyping (RCP). RTHS involves the partitioning of a reference structure into numerical and physical substructures, joined by a transfer system. In this study, the numerical substructure is a three-story, single-bay frame structure subjected to ground accelerations, while an electrodynamic proof-mass actuator (PMA) serves as both the physical substructure and the transfer system. The PMA is used to emulate an active mass damper, a tuned mass damper, and a viscous damper. The framework is validated by analyzing the tracking performance of the commanded control forces alongside the resulting attenuation of the dynamic structural responses. Ultimately, this research provides a scalable, cost-effective RCP environment. It enables highly iterative development through hardware-in-the-loop validation. This allows for the efficient testing of diverse control strategies before implementation on physical systems.

Keywords: Estimation, Assistive and Rehabilitation Robotics, Sensors and Actuators
Abstract: In rehabilitation robot systems, the estimation of user upper limb posture is one of the detection indicators to ensure the effectiveness and safety of rehabilitation, and it is also one of the bases for evaluating user rehabilitation status. Currently, the estimation of user upper limb posture in upper limb rehabilitation robots is mostly obtained directly by mapping the joint angles or end effector positions of the robot. This method assumes that the physical connection position between the robot body and the user does not change, i.e., a strong physical connection. However, in practical use, this assumption often cannot be met. In weakly physically connected rehabilitation robots, the physical connection position between the robot body and the user may change, and the binding relationship between the robot and the user in strongly physically connected robots may loosen. This can also lead to changes in the physical connection position between the robot body and the user, and the direct mapping of upper limb posture estimation can cause significant errors. Therefore, through principle analysis, a method and acceleration algorithm for estimating user upper limb posture in weakly physically connected rehabilitation robots are proposed. Through simulation experiments on the evaluation process, the proposed posture estimation method is verified. Without input interference, the estimation RMSE of shoulder flexion/extension, adduction/ abduction, internal/external rotation and elbow flexion/extension are 3.13×10-13, 8.16×10-15, 1.93×10-13 and 1.22×10-12 respectively. And if input with gaussian noise interference, the estimation RMSE of shoulder flexion/extension, adduction/ abduction, internal/external rotation and elbow flexion/extension are 0.185, 0.346, 2.06 and 0.191 respectively. The results also show that the proposed method can achieve realtime estimation of user upper limb posture under the condition of weak physical connection between the rehabilitation robot and the user.

Keywords: Intelligent Autonomous Vehicles, Path Planning and Motion Control, Unmanned Ground and Aerial Vehicles
Abstract: In Robotics and Automated vehicle applications, way-point-based motion planning (MP) is usually treated as a geometric steering method design via different kinds of curves conforming the non-holonomic kinematics. However, these methods predominantly rely on the assumption of complete state observability, particularly precise localization to trigger curvature changes. Conversely, industrial Automated Guided Vehicles (AGVs) utilizing magnetic, visual guidance, or 1D Range Sensor often operate under rank-deficient localization (i.e., accessible lateral displacement/error) and MP methods for such sensor-limited system navigation are very limited. This paper fills the gap by proposing the application of special spiral curves featuring a linear relationship between lateral coordinate and the path curvature which accounts for precise navigation even without odometer positioning. The feasibility of applying the proposed method on motion planning is verified through real-vehicle tests.

Keywords: Power and Energy Systems, Marine Systems, Control Applications
Abstract: This paper presents a nonlinear model predictive control (NMPC) for motion regulation and oscillation suppression of a dual-rotor ocean current turbine under turbulence conditions. The considered turbine exhibits strongly coupled nonlinear dynamics, in which turbulent inflow directly affects the depth response, pitch motion, and rotor angular velocities. To regulate these responses, the proposed controller manipulates the water volumes in the front and rear chambers of the buoyancy tank and computes control actions in a receding-horizon manner subject to system dynamics and operating constraints. The control objective is to maintain the desired operating conditions while suppressing turbulence-induced oscillations in the OCT depth, pitch angle, and port and starboard rotor angular velocities, without sacrificing satisfactory power-generation performance. By explicitly incorporating predictive closed-loop optimization, the proposed NMPC improves the system dynamic response and enhances its robustness under typical Gulf Stream turbulence conditions. Simulation results demonstrate improved dynamic stability, effective disturbance rejection, and reliable closed-loop operation, indicating that the proposed approach provides a promising control solution for practical deployment of dual-rotor ocean current turbines in realistic ocean environments.

Keywords: Robotics, Path Planning and Motion Control, Adaptive and Learning Systems
Abstract: We present the Agentic Robotic Additive Manufacturing Planner (ARAMP), an agentic AI framework that couples natural-language interaction with traceable, tool-grounded decision-making for robotic additive manufacturing. Instead of delegating process-critical decisions to a large language model (LLM), ARAMP uses the LLM only as a front-end for intent understanding and high-level reasoning, while specialized agents and verifiable expert tools execute the quantitative planning. The agentic workflow links slicer-generated toolpaths with workspace-aware placement, kinematic screening, velocity-profile generation, joint-motion evaluation, and synchronized extrusion, enabling joint optimization of infill pattern, orientation, and placement rather than decoupled slicer- and robot-side decisions. On a Dogbone Type IV case study, we compare a placement-only expert query with a goal-only query that jointly optimizes infill, orientation, and placement. The selected plans reduce maximum and mean absolute sixth-joint angular jerk by 10.4% and 18.3% (expert-specified) and by 48.9% and 61.0% (goal-only) relative to the least favorable valid candidate. Experimental results illustrate the potential of ARAMP as a traceable and extensible agentic planning framework for robotic AM.

Keywords: Optimal Control, Adaptive and Learning Systems, Machine Learning in modeling, estimation, and control
Abstract: In this paper, a novel bias-compensating framework is proposed to address estimation bias in learning-based adaptive dynamic programming for multiplayer dynamic games. The study focuses on linear-quadratic non-zero-sum differential games subject to unknown disturbances, where each agent optimizes its own performance objective under a single environment (game). Integral control alone does not fully remove bias arising from unmeasured disturbances, which can distort the coupled algebraic Riccati equation solutions, disrupt learning convergence, and potentially destabilize the closed-loop system. To overcome this limitation, a modified adaptive dynamic programming (ADP) formulation is presented that explicitly accounts for disturbance-induced bias through an additional aggregated term estimated alongside the control policies, enabling compensation without direct disturbance measurements. Building on this formulation, a policy iteration-based algorithm learns the Nash equilibrium policies. Simulation results confirm effective bias removal and convergence to the optimal solution.

Keywords: Optimal Control, Adaptive and Learning Systems, Cyber physical systems
Abstract: This paper develops a safety-shielded online Single-Network Adaptive Critic (SNAC) method for discounted continuous-time optimal control. A value-iteration learning law minimizes a normalized Bellman residual and computes the greedy policy directly from one critic, avoiding a separate actor network and an initial stabilizing policy. The learned policy is deployed through a Control Barrier Function Quadratic Program (CBF-QP) that minimally modifies the nominal input while preserving forward invariance of the safe set.

Keywords: Adaptive and Learning Systems, Linear Control Systems, Optimal Control
Abstract: When a controller is designed from an identified model, its performance ultimately depends on the trajectories used for identification, yet pinpointing which ones help or hurt remains an open problem. We bring influence functions, a data-attribution tool from machine learning, into this setting by chaining two closed-form sensitivity analyses across a regularized least-squares identification and an infinite-horizon linear quadratic regulator (LQR) pipeline. On the identification side, the quadratic loss admits an exact leave-one-trajectory-out (LOTO) parameter shift; a reusable first-order approximation follows with a Neumann-series error bound. On the control side, we implicitly differentiate through the discrete algebraic Riccati equation (DARE) via its discrete Lyapunov structure and compress the cost gradient to a single adjoint Lyapunov solve. The resulting scores track true LOTO retraining with Pearson correlations above 0.99, at 7× to 60× the speed, on linear systems of dimension 2 to 10, while a nonlinear benchmark confines the method to nominal, surrogate-level attribution.

Keywords: Optimal Control, Cyber physical systems, Adaptive and Learning Systems
Abstract: This paper develops a partially model-free policy-iteration framework for near-optimal tracking of continuous-time control-affine nonlinear systems. The input-gain matrix is known, whereas the drift dynamics are learned online from input--output data. A neural-network identifier estimates the nominal drift, and a Gaussian process provides a state-dependent residual-error bound. A single network adaptive critic directly approximates the optimal costate and constructs the policy. Its parameters are trained using a stabilized recursive least-squares law. Under persistence of excitation and bounded approximation errors, the tracking, critic-weight, and identifier-weight errors are uniformly ultimately bounded during and after training.

Keywords: Optimal Control, Machine Learning in modeling, estimation, and control, Adaptive and Learning Systems
Abstract: We consider data-enabled predictive control for systems operating across multiple contexts. We propose a context-aware DeePC approach that uses context-specific offline data and selects the corresponding Hankel matrices based on the detected context, while accommodating the initialization mismatch caused by context transitions. The method is evaluated on trajectory tracking of a mass-spring damper system with varying damping coefficient and a quadrotor operating in a variable wind field. Compared with standard single-context and mosaic Hankel matrix based DeePC approaches, the proposed algorithm achieves significantly improved tracking performance under changing contexts.

Keywords: Motion and Vibration Control, Adaptive and Learning Systems
Abstract: This paper presents an adaptive servomechanism control framework for rejecting sinusoidal disturbances with unknown frequency in dynamic systems. Two adaptive estimation approaches are introduced: a delay-operator-based estimation and a Lyapunov-based observer method. The proposed methods enable online frequency estimation and improved disturbance rejection compared to fixed-frequency servo-compensator designs. Simulation results on a linear system demonstrate that adaptive estimation significantly reduces tracking error. The delay-based method achieves faster convergence with increasing gain, while the observer-based approach shows sensitivity to parameter tuning.

Keywords: Path Planning and Motion Control, Human-Machine and Human-Robot Systems, Robotics
Abstract: Autonomous mobile robots in shared industrial workspaces should navigate safely around human workers while choosing an efficient route. Existing methods use short-term human path prediction for local human avoidance, but the global path planner is either prediction-unaware or relies on a single deterministic human future. We present a human-aware navigation framework consisting of human localization, state estimation, human intent prediction, and a prediction-aware global path planner. The planner uses multi-modal future human routes to construct time-dependent edge costs for global path search. We validate the proposed framework through lab experiments and a simulation study. The results show that our proposed global planner improves navigation safety by achieving a lower near-miss rate and a larger human-robot distance.

Keywords: Path Planning and Motion Control, Machine Learning in modeling, estimation, and control, Estimation
Abstract: This paper addresses the safety-critical Model Predictive Control (MPC)-based motion planning problem in the belief space under uncertainty. It first formulates a discrete-time Belief-Space Control Barrier Function (BS-CBF) to be used within a Belief-Space Model Predictive Control (BS-MPC) framework. Belief-MDP and belief-space MPC are further fused to establish the existence of a modification for a BS-MPC when the underlying belief model is imperfect. The proposed safety-critical BS-MPC relies on an Extended Kalman Filter (EKF)-based Gaussian belief model, which is an approximate and potentially imperfect representation of the true system. To tackle the belief model imperfection, a parameterized EKF-based belief model is then introduced within a parameterized safety-critical BS-MPC scheme. Finally, Multi-Objective Bayesian Optimization (MOBO) is leveraged to formulate a policy gradient-enhanced Bayesian Optimization (BO) framework for adjusting the entire parameterized framework with the aim of improving MPC-based motion planning under uncertainty. We demonstrate the efficacy of the proposed method through two representative unmanned vehicle applications: uncertain motion planning for an Unmanned Ground Vehicle (UGV) using a Wheeled Mobile Robot (WMR) model, and maritime motion planning for an Autonomous Surface Vessel (ASV). These platforms are well-established benchmarks for evaluating motion planning algorithms. Together, they demonstrate the practical relevance of the proposed approach while providing representative validation of its performance across diverse unmanned vehicle applications.

Keywords: Path Planning and Motion Control, Machine Learning in modeling, estimation, and control, Linear Control Systems
Abstract: This paper presents texttt{DaSP-LLM}, a data-driven safe motion planning approach that combines Large Language Models (LLMs) for spatially-aware waypoint generation with performance-aware invariant sets for guaranteed safety and optimality. While LLMs excel at high-level reasoning, due to their unpredictable nature, they may not provide a reachable waypoint at all times. Our framework addresses this through a feedback-driven validation process where LLM-generated waypoints are verified against data-driven safety criteria, including ellipsoidal containment, collision avoidance, and dynamic feasibility. When waypoints fail validation, the system provides specific feedback about violations and adaptively constrains the search space, enabling iterative refinement through a few-shot learning framework. This validation-feedback loop provides a link between LLM spatial-awareness and safety verification, maintaining semantic understanding while ensuring mathematical guarantees for safe execution.

Keywords: Path Planning and Motion Control, Robotics, Optimal Control
Abstract: Navigating doorways is a fundamental capability for mobile manipulators operating in human environments, requiring coordinated motion between the mobile base and manipulator arm. This paper presents a motion planning framework that generates dynamically feasible and collision-free trajectories for autonomously opening and traversing both push and pull doors. The proposed method formulates the robot and door as a coupled dynamical system within a nonlinear Model Predictive Control (MPC) optimization framework. Manipulation feasibility is enforced through a penalty-based constraint, avoiding explicit arm kinematic modeling in the planner. Simulations and a hardware experiment demonstrate that the approach successfully plans feasible trajectories for door traversal.

Keywords: Robotics, Optimal Control, Path Planning and Motion Control
Abstract: This paper presents a real-time cooperative nonlinear model predictive control (NMPC) framework for coordinated humanoid upper-body motion, jointly optimizing both arms and the torso within a unified predictive structure. A lightweight custom NMPC implementation is developed from first principles, enabling fine-grained control of algorithmic parameters, facilitating deployment on embedded humanoid platforms, and exposing explicit trade-offs between accuracy and computational cost. The controller enforces joint limits, self-collision avoidance, and cooperative task objectives, allowing the torso to actively enhance reachability, manipulability, and task feasibility in constrained environments. To further mitigate computational burden, a neural network is trained as a high-fidelity surrogate to approximate NMPC trajectories, validating its potential to provide efficient warm-start initialization for the solver with negligible evaluation time. Statistical profiling shows that, while the solver exhibits stable convergence, closed-loop execution without acceleration is insufficient for real-time operation. The learned warm-start significantly improves convergence speed while preserving robustness. Simulation and physical experiments demonstrate smooth, feasible, and collision-free arm–torso coordination in both static and dynamic scenarios. Results confirm that torso involvement expands the effective workspace and improves task robustness, while learning-assisted acceleration enables practical real-time execution. The proposed framework provides a scalable pathway toward full upper-body predictive control in humanoid robots.

Keywords: Robotics, Path Planning and Motion Control, Nonlinear Control Systems
Abstract: Most planners focus on obstacle avoidance, but legged robots can step over obstacles, reducing travel distance and time. We address this gap by integrating motion control and planning within a sampling-based planner. We develop a constrained inverse-dynamics controller with task-space feedback that tracks center-of-mass velocities via a closed-form solution, thereby avoiding iterative optimization. This enables the use of reduced-order planning models by ensuring dynamic consistency at the full-body level. For planning, we introduce BiKE-Legs and DRiVE-Legs, which augment classical bicycle and differential-drive kinematics with leg-based stepping capabilities. These models are embedded within the Dynamic Window Approach (DWA), with cost functions that balance goal progress, obstacle negotiation, and speed to enable selection between avoidance and stepping. Results across scenarios with small and large obstacles using reduced kinematic models, along with limited validation on a simulated Digit humanoid, demonstrate the effectiveness of the proposed approach.

Keywords: Assistive and Rehabilitation Robotics, Human-Machine and Human-Robot Systems, Control Design
Abstract: Rehabilitation robots support interactive movement training after neurological injury but are often highly specialized, limiting scalability. Industrial collaborative robots (cobots) offer a potential alternative due to their widespread availability and standard platforms, but their suitability for rendering transparent environments necessary for rehabilitation remains unclear despite increasing attention in recent work. This work presents a preliminary simulation framework that couples a cobot with a simplified 7-DOF human arm model to evaluate admittance-controlled interaction. Dynamic interactions across three trajectories provide initial insight into the haptic performance of cobots for upper-extremity rehabilitation tasks

Keywords: Biomechanical Systems, Estimation
Abstract: Several neurological movement disorders are characterized by unwanted, chaotic, involuntary tremors that vary in severity as well as location. Often stemming from neurological impairment, these movement disorders, such as Essential Tremor, Parkinson’s Disease, and Dyskinesia, lead to involuntary movements that can negatively impact a person’s quality of life. As a result, neurologists have developed rating scales for clinicians to rate tremor severity and standardize impairment assessment. However, characteristics and causes of these tremors vary by the underlying disease, leading to multiple severity scales being developed such as the Fahn-Tolosa-Marin Clinical Rating Scale and the Essential Tremor Rating Assessment Scale. One commonality among these varying scales is the use of handwriting and the drawing of an Archimedes Spiral to aid in diagnosing tremor severity in the hands. However, the writing and drawing portions of these assessments often rely on the clinician assessing the tremor severity based on subjective metrics outlined in the official scales. Additionally, these assessments are done by drawing on a table, allowing for the participant being assessed to steady their hand by pressing down on the table as it provides a near infinite stiffness. Potentially masking underlying compensatory tactics, we hypothesize that Archimedes Spiral based assessments may mask tremor characteristics that could be captured in three dimensions. To this aim, we implement an Archimedes Spiral task with 8 healthy participants and 3 participants affected by neurological movement disorders with the 3D Systems Phantom Touch haptic device. The assessment protocol included a high impedance physical surface and a low impedance haptic floor for the drawing task. Results indicate healthy individuals move smoother with traditional high impedance spiral drawing, while individuals affected by tremors may write smoother under a lower impedance environment, which promotes future investigation into the understanding of movement quality and tremor classification in three-dimensional space.

Keywords: Healthcare systems, Estimation, Biomechanical Systems
Abstract: This paper develops a method for wearable tremor quantification using a finger-mounted inertial measurement unit (IMU) sensor. Sensor orientation is estimated using a quaternion-based complementary filter and then used for gravity removal prior to displacement estimation. Tremor magnitude is then obtained as a physically meaningful displacement amplitude from the IMU signal through frequency-domain power spectral density (PSD) analysis within the tremor band. The method is evaluated against a laboratory-grade ground truth reference system. Results show improved agreement with the reference compared with a raw acceleration plus bandpass method, and also indicate that frequency-domain analysis is more robust than direct time-domain displacement estimation for a 6-axis IMU. The proposed framework supports practical wearable assessment of tremor in Parkinson’s disease.

Keywords: Healthcare systems, Modelling, Identification and Signal Processing, Control Applications
Abstract: Sustaining behavior change in digital health interventions is often hindered by over-reliance on external rewards, as fixed reinforcement schedules fail to induce self-sustained long-term physical activity among participants. To address this limitation, this paper proposes a data-driven modeling and control framework that designs reinforcement strategies capable of reducing reward dependence while preserving engagement. Participant behavior is first identified using a Model-on-Demand (MoD) estimator, which generates locally linear models from sparse and noisy data. A Discrete Simultaneous Perturbation Stochastic Approximation (DSPSA) algorithm guides feature and regressor selection, yielding parsimonious yet expressive personalized models. At each prediction step, MoD constructs a local model from the participant’s historical estimation data, providing a participant-specific surrogate for closed-loop simulation. These models are subjected to a Three-Degree-of-Freedom Kalman Filter-based Hybrid Model Predictive Control (3DoF-KF HMPC) formulation relying on Mixed Logical Dynamical (MLD) framework, which integrates logical reinforcement rules, process constraints, and randomized point-independent phases. The resulting closed-loop system optimally balances goal-setting and incentive delivery while fostering self-reinforcing behaviors. Case studies with participant data from the YourMove intervention demonstrate that the proposed MoD–DSPSA and 3DoF-KF HMPC framework maintains behavioral performance during intermittent removal of rewards, offering a principled and scalable approach to adaptive intervention design.

Keywords: Machine Learning in modeling, estimation, and control, Healthcare systems, Control Applications
Abstract: "Just-in-Time" adaptive interventions (JITAIs) can provide daily step goals and financial incentives (e.g., reward points) to enable participants to reach healthy levels of physical activity. However, personalization and optimization remain challenging because participant responses vary over time and between individuals. Recent JITAI efforts have benefited from control-oriented strategies using system identification and Model Predictive Control (MPC), with YourMove (NCT05598996) providing an illustrative example. MPC offers a structured decision framework for selecting feasible goal trajectories from theory-guided dynamic behavioral models, but its closed-loop performance depends on tuning choices that trade off tracking speed, smoothness, constraint satisfaction, and other criteria, making manual selection difficult. A contextual multi-armed bandit formulation can instead use a reward function to select MPC tuning parameters that best align the controller with the intervention outcomes of interest, conditioned on observed participant context. This paper presents a simulation-based proof of concept in which a TD3-based agent is used as an outer-loop tuner that learns the optimal MPC tuning parameters, while MPC produces feasible responses that satisfy intervention constraints. Results demonstrate improved closed-loop behavior tracking and increased self-efficacy relative to a fixed-tuning MPC baseline, and multi-seed training indicates reproducible convergence to high-reward tuning policies. Both novel and practical insights on the synergism of machine learning and control engineering are outcomes of this research.

Keywords: Modelling and Control of Biomedical Systems, Assistive and Rehabilitation Robotics, Nonlinear Control Systems
Abstract: The quadriceps and knee joint are a major rehabilitation target for individuals with neuromuscular conditions in order to improve gait performance. Hybrid devices that provide functional electrical stimulation (FES) and motor actuation are a common tool used to augment a user's natural ability, but rapid onset of FES-driven fatigue limits their application. In this work, we developed a novel wearable ultrasound (WUS)-based algorithm for providing real-time feedback to a fatigue-aware Koopman model predictive controller (FA-KMPC) for hybrid knee extension. We performed simulations and experiments which demonstrate that the FA-KMPC fully utilizes FES when fatigue is low and reduces FES with increasing fatigue. During experiments, the FA-KMPC resulted in a notably lower knee angle tracking NRMSE of 10.14% compared to the 19.22% of a KMPC with no fatigue feedback. Our results demonstrate the feasibility of our approach, and we plan to further robustify our WUS-based fatigue feedback and expand application of FA-KMPC to multi-muscle control.