Keywords:
Abstract: For decades, the field of control systems has focused on the modeling, analysis, and design of physical dynamical systems arising in diverse engineering domains, including aerospace, automotive, energy, healthcare, manufacturing, robotics, and transportation.

This has led to a rich set of tools centered on performance, safety, prediction, optimization, security, and robustness, largely within the context of engineered systems.

One of the messages articulated in the recent CSS publication, “Control for Societal-scale Challenges: Road Map 2030,” is that control systems should be viewed not only as a mathematical framework for analyzing and synthesizing dynamical systems, but also as a systems-level methodology for addressing complex societal challenges. While achieving system level objectives in engineering applications has undoubtedly had indirect benefits for individuals and society at large, this talk argues for moving beyond such borders.

The talk explores how the core principles and tools of control can be applied directly to societal systems, with explicit consideration of human-centered objectives. Two case studies will illustrate this perspective: the design of an affordable microtransit service and a resource management framework aimed at advancing energy justice. The talk will also engage with broader societal concerns—including equity, fairness, and affordability—highlighting the opportunities and responsibilities for the control community in shaping systems that directly serve society.

Keywords: Fault detection/accomodation, Energy Storage, Observers
Abstract: Accurate fault diagnosis is essential for enhancing the safety and reliability of lithium-ion batteries. This study focuses on the detection, estimation, and isolation of two major degradation-related electrochemical faults: Solid Electrolyte Interphase (SEI) film growth and lithium plating. The Single Particle Model with electrolyte dynamics augmented by Side Reactions (SPMe+SR), incorporated through modified boundary conditions, serves as the foundation for the proposed fault diagnosis scheme. Cascaded Partial Differential Equation (PDE) backstepping observers are designed for both the electrolyte and solid phases. The first observer, referred to as the state observer, estimates the distributed lithium concentration. A residual is then constructed for fault detection. Upon detecting a fault, the second observer, termed the fault estimator, employs the concentration estimated by the state observer to quantify the magnitudes of SEI film growth and lithium plating. Furthermore, the fault magnitude estimates are exploited to develop a fault isolation scheme that identifies the fault origin. The effectiveness of the proposed framework is validated through simulation.

Keywords: Biologically-inspired methods, Intelligent systems, Machine learning
Abstract: The 2022 re-emergence of Monkeypox (Mpox) as a Public Health Emergency of International Concern underscores the critical need for precise temporal forecasting to optimize medical resource allocation and outbreak control strategies. Although the global case fatality rate remains around 0.7%, regional vulnerabilities present highly non-linear and heterogeneous mortality risks. Tradi-tional sequence models often struggle to balance long-range dependencies with local temporal variations under restricted datasets. To solve this limitation, this study introduces TRAGIC, a novel hybrid deep learning architecture that fuses the global context-awareness of Transformers with the localized sequential processing of Gated Recurrent Units (GRUs). Crucially, a lightweight Quick Attention mechanism is integrated to dynamically weigh high-impact clinical and demographic features while maintaining computational efficiency. Benchmarked against benchmark LSTM, GRU, and standard Transformer models using a comprehensive global dataset of 16 multi-dimensional features, TRAGIC achieves superior predictive precision. The findings demonstrate that this ad-vanced fusion framework serves as a robust, interpretable decision-support tool, enabling public health authorities to implement proactive, data-driven interventions and optimal dynamic control of medical logistics.

Keywords: Distributed control, Adaptive control, Robotics applications
Abstract: Bearing-based formation control has attracted considerable attention for multi-agent systems since bearing measurements require minimal sensor capability and can be obtained with inexpensive vision-based or directional sensors. For infinitesimally bearing rigid formation systems, bearing-based control laws are known to guarantee global stability in the disturbance-free setting. In our earlier work, an adaptive bearing-based control law was proposed, where the control gain of each agent is dynamically adjusted according to the real-time bearing formation error, and global stability was established. In practice, however, various sources of exogenous disturbances exist, including sensor noise, device aging, and wind gusts. Numerical simulations demonstrated that the adaptive law achieves faster convergence and smaller steady-state errors than a conventional fixed-gain law, suggesting an inherent robustness advantage. Despite this observed improvement, a rigorous theoretical characterization of the robustness properties in the presence of exogenous disturbances has not yet been established. In this work, we address this gap by extending the framework to single-integrator dynamics with bounded time-varying exogenous disturbances. Via a Lyapunov stability analysis, we show that the bearing formation error locally asymptotically converges to an explicit upper-boundary set parameterized by the disturbance bound, the eigenvalues of the bearing rigidity matrix, and the initial control gain. The set is valid provided the disturbance magnitude does not exceed a threshold determined by the initial gain and the bearing rigidity eigenvalues. Furthermore, when the initial control gains satisfy ki0 > 1 for all agents, the adaptive mechanism provides two robustness advantages over the conventional non-adaptive law: the allowable disturbance bound is enlarged by a factor of the minimum initial gain, and the upper-boundary set is strictly smaller under the same disturbance level.

Keywords: Stochastic/uncertain systems, Chaotic systems, Nonlinear systems
Abstract: Designing robust interventions for high-dimensional and nonlinear systems remains a fundamental challenge, particularly when forecast uncertainty limits reliability of deterministic optimization. We propose a framework that reinterprets forecast uncertainty as an informative resource for intervention design. The framework adopts an ensemble-based predictive control strategy in which ensemble predictions represent future uncertainty and state-dependent sensitivity, approximating the mapping from interventions to future outcomes. The framework extracts a representative ensemble trajectory and designs control inputs within a feedback loop informed by data assimilation. By relying on ensemble-based approximations, this approach avoids the need for tangent linear or adjoint models and is applicable to computationally demanding numerical simulations. We apply the framework to an extreme precipitation event in Japan in August 2021 using a convection-permitting atmospheric model. Interventions are parameterized as idealized frictional forcing over an oceanic region, and their impact is evaluated based on precipitation over a downstream land area. The proposed approach achieves an approximately 23% reduction in accumulated precipitation over the target region on average across independent realizations, demonstrating robustness under realistic forecast uncertainty. These results indicate that effective intervention design is possible without perfect predictability, and highlight the potential of using forecast uncertainty for decision-making in complex Earth systems.

Keywords: Simulation, Sensor fusion, Predictive control
Abstract: This poster proposes a real-time state estimation and trajectory monitoring framework based on AI vision recognition for the safe control and automation of ground vehicles in a complex and dynamic airport airside environment. With the increasing scale of airport operation vehicles, securing highly reliable feedback data is essential for collision avoidance and precise tracking control. However, conventional GPS-based positioning systems suffer from limitations in continuous vehicle state tracking due to signal shadow zones and multipath errors caused by large airport structures. To address these challenges, this study develops a high-precision virtual sensor system that leverages existing airport CCTV infrastructure to substitute or augment physical sensors. The proposed system extracts vehicle bounding boxes using deep learning-based object detection and multi-object tracking algorithms, which are then mapped onto an airport digital twin coordinate system to estimate dynamic state variables such as real-time position, and velocity. In particular, to reject visual noise and uncertainties arising from adverse weather conditions (e.g., night, fog), a state observer based on the Extended Kalman Filter is integrated with the AI analytics, thereby ensuring the robustness of trajectory estimation against disturbances. Experimental validation conducted under operational scenarios modeled after Incheon International Airport demonstrates that the proposed vision-based state estimation system satisfies the real-time feasibility required within a control loop while providing stable feedback signals despite external disturbances. This research signifies that AI vision analysis can expand beyond the computer vision domain to function as a reliable input sensor for model predictive control and collision avoidance algorithms in future airport ground traffic management.

Keywords: Robotics applications, Control architectures, Mobile Robots
Abstract: The increasing deployment of heterogeneous mobile robots in complex public facilities calls for an integrated supervision framework that can coordinate robots from multiple vendors under unified operational and safety requirements. However, practical deployment remains challenging due to differences in robot middleware, communication protocols, data structures, map representations, and vendor-specific control interfaces. This work presents the development of an integrated supervision and interoperable control framework for heterogeneous mobile robots, derived from an operational guideline study for multi-robot deployment in airport-scale environments. The proposed framework consists of a four-layer architecture spanning robot, edge, platform, and operation/service layers, together with a common data model (CDM) for standardized representation of robot identity, pose, battery status, mission progress, telemetry, and safety states. To support cross-vendor interoperability, the framework incorporates protocol mediation across DDS, MQTT, and API-based interfaces, as well as orchestration functions for mission allocation, state monitoring, shared-resource coordination, and safety event handling. In addition, quantitative validation targets are defined for communication latency, command response, localization accuracy, and safety distance, and a digital-twin-based verification pipeline is outlined for functional, performance, and integrated operational assessment. The poster highlights the architecture development, interoperable information model, and preliminary validation perspective toward scalable integrated control of heterogeneous robots in real-world facilities.

Keywords: Simulation, Reinforcement learning, Robotics
Abstract: Sim2real robot learning is often hindered by fragmented workflows across simulation, control, training, and validation, which limits reproducibility and reliable deployment. We present VTPRL, a unified platform for virtual training and evaluation of robust and continual reinforcement learning policies. VTPRL integrates digital twins, learning agents, classical controllers, monitoring, replay, and profiling within a modular pipeline, enabling structured domain randomization, repeatable scenario-based testing, and scalable parallel rollouts. The platform supports the full lifecycle from pre-deployment policy training to sim2real transfer, runtime performance monitoring, and post-deployment adaptation under changing conditions. As the common backbone for multiple sim2real studies, VTPRL provides a control-oriented infrastructure for systematic benchmarking, validation, and deployment of learning-enabled robotic systems.

Keywords: Automotive applications, Mechatronic systems
Abstract: Battery electric vehicles play an important role in personal transportation and are becoming more popular in motorsports. This work investigates the performance impact of jointly optimizing cooling control inputs and cooling system topology for high-performance battery-electric vehicles operated under motorsport conditions. The problem is formulated as a nonlinear minimum-lap-time optimal control problem with thermal dynamics and temperature constraints, and co-optimize the cooling actuation and flow allocation (topology). The problem is implemented in CasADi and solved using IPOPT. Results for the Monaco circuit show that optimizing the cooling topology can reduce the total time over four consecutive laps by up to 25s by enabling a more effective allocation of cooling capacity to thermally limiting components.

Keywords: Automotive applications, Control applications
Abstract: This study proposes a clutch kissing point estimation method for vehicles equipped with a Ball-Ramp Dual Clutch Transmission (BR-DCT) during the power-on downshift process. The BR-DCT utilized in this research incorporates a self-energizing principle, characterized by an asymmetric torque capacity in which forward torque transmission is significantly higher than inverse transmission for the same actuator force. This asymmetry limits the delivery of inverse torque, thereby mitigating the degradation of shift quality typically caused by "tie-up". Utilizing this characteristic, we introduce a strategy to detect the on-coming clutch kissing point by proactively initiating engagement during the Inertia Phase (IP), prior to the start of the dedicated downshift clutch torque control. This approach reduces input errors caused by kissing-point uncertainties during the Torque Phase (TP), ensuring that shift control quality remains consistent with the original design. The feasibility of the proposed detection method was validated using experimental equipment replicating an actual DCT, and the control advantages were demonstrated through comprehensive simulations.

Keywords: Automotive applications, Data analytics, Control Technology
Abstract: Driver assistance systems are expected to interact with human drivers in a natural and intuitive manner while supporting passenger comfort and safety in everyday traffic situations. However, existing implementations can exhibit artificial or non-human-like behavior, which challenges user acceptance. Such systems are often tuned manually by experts during test drives and rely on limited reference drives; therefore, they may fail to capture the variability and context dependence of human driving behavior. This paper presents a data-driven approach to learn offline behavior-consistent control parameters from real-world driving data using a simple, interpretable controller structure. Human driving behavior is represented implicitly through structured regions in the parameter space of a parameterized control law, identified from observed driving trajectories. The proposed concept is demonstrated using adaptive cruise control as a representative application. First, control parameters are learned from large-scale real-world driving data, showing that a simple and interpretable control structure can reproduce a wide range of human approach behaviors. Second, a transformation-based approach is introduced to transfer learned behavior-consistent parameterizations between different driving scenarios, exemplified by the transformation from slow to fast approach maneuvers. While demonstrated for adaptive cruise control, the approach is generalizable in principle to other driver assistance functions and contextual variations. Overall, the results illustrate how learning behavior-consistent control parameters from real-world data can help bridge the gap between human driving behavior and automated driver assistance systems without relying on black-box models.

Keywords: Automotive applications, Estimation, Learning
Abstract: This paper proposes a hybrid estimator that predicts the peak friction coefficient under combined slip conditions by leveraging a physically-based preprocessing stage with a neural network classifier. By embedding physical knowledge and exploiting classification scores with heuristics, the method reliably identify the most likely grip class even under low excitation. The selected class is then mapped to a nominal peak friction value. Experimental results on throttle-on-exit manoeuvres across different surfaces demonstrate the algorithm accuracy and predictive capability.

Keywords: Automotive applications, Fault detection/accomodation, Kalman filtering
Abstract: Sensor fault detection is crucial for safe vehicle control, requiring rapid and accurate estimation under highly nonlinear dynamics. While the unscented Kalman filter (UKF) provides a powerful framework, augmenting the state vector with faults and parameters significantly increases computational load. To address this, this paper proposes a numerically efficient, fault-tolerant vehicle state and road friction estimation architecture based on an adapted two-stage square-root spherical simplex UKF (TS3UKF) formulation. First, algorithmic optimization is achieved by implementing the spherical simplex method - which reduces the standard 2n+1 sigma points to n+1 points - within the two-stage architecture while directly updating Cholesky factors to guarantee numerical stability. Second, the system state dimension is minimized by integrating an efficient tire slip calculation scheme that avoids full sigma point propagation. Finally, the framework is experimentally verified under highly critical, low-friction conditions using an Audi SQ8 e-tron test vehicle. Real-world results demonstrate that the algorithm precisely tracks states under injected IMU sensor faults with minor delay. In the absence of sensor faults, the road friction is estimated with high accuracy. Under active sensor faults during steady-state limit cornering, the friction estimate remains safely bounded, whereas subsequent dynamic excitation significantly improves convergence accuracy. Consequently, the total number of nonlinear function evaluations is reduced by 62,%, proving the modified algorithm's high suitability for production vehicles.

Keywords: Autonomous systems, Automotive applications, Planning
Abstract: Modular vehicles present a novel area of academic and industrial interest in the field of multi-agent systems. Modularity allows vehicles to connect and disconnect with each other mid-transit which provides a balance between efficiency and flexibility when solving complex and large scale tasks in urban or aerial transportation. This paper details a generalized scheme to route multiple modular agents on a graph to a predetermined set of target nodes. The objective is to visit all target nodes while incurring minimum resource expenditure. Agents that are joined together will incur the equivalent cost of a single agent, which is motivated by the logistical benefits of traffic reduction and increased fuel efficiency. To solve this problem, we introduce a novel algorithm that seeks to balance the optimality of the path that every single module takes and the cost benefit of joining modules. Our approach models the agents and targets as point charges, where the modules take the path of highest attractive force from its target node and neighboring agents. We validate our approach by simulating multiple modular agents along real-world transportation routes in the road network of Champaign-Urbana, Illinois, USA. The proposed method easily exceeds the available benchmarks and illustrates the benefits of modularity in multi-target planning problems.

Keywords: Autonomous systems, Automotive applications, Sensors
Abstract: High-speed multi-vehicle autonomous racing increases the safety and performance of road-going Autonomous Vehicles. Precise Vehicle Detection and Tracking from a moving platform is a key requirement for planning and executing complex autonomous overtaking maneuvers. To address this requirement, we have developed a latency-aware EKF-based Multi-Target Tracking algorithm fusing LiDAR and RADAR measurements.. The algorithm exploits the different sensor characteristics by explicitly integrating the range-rate in the EKF measurement function, as well as a priori knowledge of the racetrack during state prediction. It can handle Out-Of-Sequence Measurements via Reprocessing using a double state and measurement buffer, ensuring sensor delay compensation with no information loss. This algorithm has been implemented on Team PoliMOVE's autonomous racecar, and was validated experimentally by completing a number of fully autonomous overtaking maneuvers at speeds up to 275 km/h.

Keywords: Autonomous systems, Computational methods, Data analytics
Abstract: This paper proposes a CAN (Controller Area Network) data-based vehicle weight estimation algorithm for predicting tire wear in an electrified PBV (Purpose-Based Vehicle). Unlike dynamics-based state estimators that require accurate model parameters, the proposed method leverages CAN signals that are highly correlated with weight changes, making it practical for vehicles without dedicated weight sensors.

Keywords: Autonomous systems, Discrete event systems, Manufacturing systems
Abstract: Manufacturing systems exhibit strong concurrency, synchronization, and contention for shared reusable resources, which makes fast and reliable scheduling and verification challenging. Petri nets provide a rigorous formalism for modeling such discrete-event manufacturing systems, but reachability analysis and solving remain difficult for conventional graph search or optimization-based solvers, particularly under state-space explosion and evolving production requirements. Recently, Large language models (LLMs) have shown promise as flexible planners that can generate candidate action sequences from textual specifications. However, direct use of LLMs for Petri net reachability remains unreliable. This paper proposes an LLM-based solving framework augmented with a two-stage reflection and reprompting mechanism. The combined effects of reflection and re-clarification improve the accuracy of feasible sequence generation. The proposed method is evaluated on an industrial case modelled as a Petri net. Under a fixed Petri net structure, the proposed strategy is assessed on six solvable reachability configurations. The results demonstrate improved reliability and stability in solving Petri net reachability problems. The proposed framework is further evaluated across multiple LLMs, which indicates that the framework is not tied to any specific model.

Keywords: Autonomous systems, Intelligent systems, Reinforcement learning
Abstract: A key step toward autonomous industrial operation is the ability to create and reconfigure control policies from natural-language requirement specifications, with minimal or no manual redesign. In this setting, policy generation by AI agents can be a credible path when paired with a plant-aware validator, such as a digital twin, that can check generated candidate actions before execution. However, practical deployment is constrained by inference latency and compute footprint: large cloud-based models are often too slow, opaque, or data-sensitive for edge closed-loop use. This work investigates whether a compact Small Language Model (SLM) can be retrained for control reasoning and embedded in a validator-guided correction loop. We use a Qwen2.5-1.5B model aligned via Group Relative Policy Optimization (GRPO), combined with an action agent, a symbolic/digital-twin-style validation layer, and a reprompting agent that iteratively steers outputs toward valid actions. In randomized thermal-control simulations consisting of 30 experiments with 500 steps each, the framework achieves 91.5% average action-alignment accuracy, ranging from 86.3% to 100% across cases, at 3.84 s mean inference latency. Under symbolic re-mapping, it maintains a 95% in-range rate, indicating robust physical regulation despite reduced token-level agreement. These results support SLM+validator architectures as a practical path toward reconfigurable autonomous control at the edge.

Keywords: Autonomous systems, Planning, Automotive applications
Abstract: Autonomous valet parking (AVP) requires efficient motion planning and replanning, particularly in changing environments where parking lot occupancy is discovered on the go. Traditional search- or optimization-based approaches often struggle with real-time adaptability when integrating new occupancy information. This paper presents a motion-primitives-based planning framework that efficiently generates and updates trajectories within dynamically detected free-space corridors. To validate the proposed method, we conduct hardware-in-the-loop (HiL) testing by integrating a real camera with perception, planning, and control algorithms running on embedded hardware. This setup demonstrates the real-time feasibility of the planning within a complete AVP system operating in a closed-loop simulated environment. Experimental results show that our approach enables fast and reliable trajectory planning, navigating the vehicle to desired parking spots while respecting traffic rules, even when occupancy changes or the user selects a different spot.

Keywords: Autonomous systems, Marine/underwater robotics, Adaptive control
Abstract: Marine robots must maintain precise control and ensure safety during tasks such as navigating narrow waterways, even when they encounter unpredictable disturbances that impact performance. Designing algorithms for uncrewed surface vehicles (USVs) requires accounting for these disturbances to control the vehicle and ensure it avoids obstacles. While adaptive control has addressed USV control challenges, real-world applications are limited, and certifying USV safety amidst unexpected disturbances remains difficult. To tackle control issues, we employ a model reference adaptive controller (MRAC) to stabilize the USV along a desired trajectory. For safety certification, we developed a reachability module with a moving horizon estimator (MHE) to estimate disturbances affecting the USV. This estimate is propagated through a forward reachable set calculation, predicting future states and enabling real-time safety certification. We tested our safe autonomy pipeline on a Clearpath Heron USV in the Charles River, near MIT. Our experiments demonstrated that the USV’s MRAC controller and reachability module could adapt to disturbances like thruster failures and drag forces. The MRAC controller outperformed a proportional–integral–derivative (PID) baseline, showing a 45%–81% reduction in position root mean squared error. In addition, the reachability module provided real-time safety certification, ensuring the USV’s safety. We further validated our pipeline’s effectiveness in underway replenishment and canal scenarios.

Keywords: Energy Storage, Automotive applications, Machine learning
Abstract: Accurate state-of-health (SOH) estimation is essential for safe operation and reliable fault diagnosis of lithium-ion batteries. However, existing data-driven methods often rely on specific voltage windows or health indicators that may shift with aging, limiting their applicability when partial discharge events are common. This paper proposes a robust SOH estimation framework based on differential capacity (dQ/dV) curves for partial discharge conditions. The framework uses a reference-concatenated input representation that pairs current and initial-cycle dQ/dV curves, enabling the model to learn degradation-related electrochemical changes. To address incomplete discharge data, a history-based imputation strategy reconstructs missing curve segments using earlier cycles. An attention-based feedforward neural network (FNN-Attention) is developed that incorporates multi-head self-attention to capture complex dependencies across voltage levels in the dQ/dV curves. Numerical studies are performed using the MIT-Stanford battery aging dataset, where the proposed FNN-Attention achieves 0.38% root mean square percentage error (RMSPE), an 11.6% improvement over the state-of-the-art reported in literature. With partial discharge curves retaining only 30% (70% of the dQ/dV curve is missing) of the full curve, the FNN-Attention maintains 0.47% RMSPE using history imputation and 0.89% without requiring historical data. The results demonstrate the effectiveness and practical applicability of the proposed framework for real-world battery management systems.

Keywords: Energy Storage, Automotive applications, Modeling
Abstract: Rechargeable lithium-metal batteries (LMBs) offer the potential for substantially higher energy density than lithium-ion batteries, but their deployment in high C-rate, weight-critical applications requires accurate and efficient models suitable for embedded battery-management systems. This work presents a data-sparse physics-informed Gaussian-process (GP) model that combines an enhanced single-particle model (SPMe) with learned residual dynamics to address accuracy degradation at high C-rate. We introduce a single-pole filter to capture longitudinal electrode non-uniformity neglected by conventional SPM formulations, and leverage GP regression to fuse this added dynamic with the SPMe. To reduce memory requirements, we employ a farthest-point sampling strategy to select a compact yet informative training dataset. By construction, our model guarantees stability and reverts to the vanilla SPMe for input stimuli far enough removed from the training data. Validation against a full-order electrochemical model demonstrates significant error reduction at high C-rates, including an 86% reduction in voltage prediction error for a cell with a 200 µm-thick electrode at 1.8C, and the GP model requires only 1.6% of the storage of our previously developed physics-informed neural-network (PINN) model.

Keywords: Energy Storage, Energy Systems
Abstract: Electrochemical impedance spectroscopy (EIS) is a key tool for non-invasive battery characterisation, providing a compact representation of physical processes over many time scales. Often, EIS relies on single sinusoids excited at different frequencies sequentially, but a multisine (sum of sines) excitation is advantageous for reducing measurement time and for nonstationary tests - during charging, discharging, relaxation, and temperature changes. In this work, we measure multisine EIS on a Li-ion battery over five orders of magnitude in frequency using a commercial potentiostat and show that measurements are almost identical to single-sine tests. We then measure operando impedance with multisine at nonstationary operating conditions, first with a changing temperature, then with a non-zero DC current, and show that a family of (local) frequency responses may be extracted from one test. Finally, results are further investigated by comparison with a nonstationary impedance computation from a physical model, linearising a state-space representation around a non-zero operating point.

Keywords: Energy Storage, Neural networks, Estimation
Abstract: Reliable state-of-charge (SOC) and state-of-health (SOH) estimation is fundamental to safe and efficient battery operation. However, traditional data-driven SOC-SOH estimation methods are confined to a ``numerical-output-only'' paradigm, suffering from poor explainability and failure to convert results into actionable engineering decisions. This paper proposes a semantic-enhanced self-explainable SOC–SOH joint estimation framework built on a large-small model interactive mechanism: a deep-learning neural network (tool model) processes raw battery data to output numerical SOC-SOH estimates, while a large language model (LLM) complies these results with embedded prior knowledge via a semantic reasoning paradigm to generate explainable, reliability-aware conclusion. Concretely, the tool model adopts a dual-branch convolutional architecture to capture short-term SOC dynamics and long-horizon SOH degradation patterns for accurate joint estimation. Based on these estimates, cascaded prior knowledge mining and statistic-to-semantic conversion construct structured, domain-knowledge–rich prompts. Guided by the prompts and tool outputs, the LLM generates human-readable explanations with supporting evidence, domain-consistency checks, and reliability conclusions, closing the loop from state estimation to health insight. Effectiveness of the proposed method is demonstrated using a LIB aging case.

Keywords: Energy Systems, Energy Storage, Renewable Energy
Abstract: Stringent ramp-rate constraints due to high solar photovoltaic (PV) penetration in power systems challenge conventional smoothing techniques, where capacity saturation and device degradation become critical bottlenecks. In this context, PV smoothing is posed as a constrained dispatch objective to shape the grid-facing net power trajectory and reduce ramp-rate deviations. This study proposes a grid-compliant coordination framework for PV-integrated Hybrid Energy Systems (HES) that strategically couples battery storage with a hydrogen subsystem, comprising an electrolyzer and a fuel cell, to maximize system-wide efficiency while mitigating ramp-rate excursions, subject to operational limits of the storage devices. We develop a hierarchical supervisory control architecture consisting of an inner-loop real-time battery control and a high-level nonlinear Model Predictive Control (MPC) dispatcher. The MPC framework optimizes the trade-offs between State of Charge (SOC) regulation, conversion efficiency, device switching frequency, and ramp-rate support. Benchmarked against a logic-based control scheme, sensitivity analyses across prediction horizon, battery capacity, and forecast stochasticity demonstrate that MPC significantly improves dispatch smoothness and manages SOC effectively. The results quantify the operational flexibility contributed by hydrogen assets when battery approaches saturation, providing a robust solution for reliable grid integration of high-variability renewables.

Keywords: Energy Systems, Predictive control
Abstract: This paper proposes an economic model predictive control (MPC) framework for a heat pump in an integrated solar thermal system. This framework aims to simultaneously maintain the domestic hot water load, minimize electricity costs to operate the heat pump, and mitigate actuator fatigue. The thermal dynamics model of the solar thermal system used in MPC is built using experimental data. The MPC utilizes temperature feedback to compute the optimal power modulation ratio for the heat pump operation. The simulation over a 3-day winter period demonstrates that the proposed strategy effectively shifts energy consumption to off-peak periods via pre-heating. It achieves a 69.1% reduction in electricity costs compared to traditional rule-based control while robustly meeting the hot water load.

Keywords: Control applications, Renewable Energy, Sliding mode control
Abstract: Among the class of robust trajectory tracking controllers and observers used in wave energy systems, sliding mode (SM) algorithms are an attractive alternative. While theoretically sound, high-frequency oscillations and restrictive design assumptions appearing in practice are typical criticisms towards classic SMs. However, recent theoretical advances in SM algorithms force revisiting these strategies as potential tools for robust control, and observation, of wave energy systems. Specifically, this paper revisits, from a wave energy application perspective, filtering structures that enhance noise resilience, higher-order control strategies that synthesise smooth control actions, and nesting strategies that relax restrictive relative degree conditions on the sliding variable, permitting the design of robust and smooth position-tracking~ac{SM} controllers for wave energy systems.

Keywords: Linear robust control, Predictive control, Renewable Energy
Abstract: This paper investigates the control of wave energy converter (WEC) arrays with strong hydrodynamic coupling. While model predictive control (MPC) is effective for constrained energy optimization, centralized MPC becomes computationally impractical for large arrays. This work studies a robust control framework, built upon a pseudo-decentralized MPC (PD-MPC) architecture, for operation in a wide range of sea states. In the PD-MPC framework, each WEC employs a local controller that solves a reduced-order MPC problem using local measurements and limited information exchange with neighboring devices. This structure reduces computational and communication demands while respecting radiation coupling effects, enabling coordinated management of hydrodynamic interactions. Robust MPC is incorporated to explicitly address uncertainties in modeling errors, ensuring constraint satisfaction and stable performance under imperfect wave information. The framework is evaluated on a four-float WEC array across three representative sea states, selected to reflect realistic offshore operating conditions, and is shown to achieve safe operation with strong energy capture performance.

Keywords: Renewable Energy, Control applications, Observers
Abstract: Energy-maximising control of wave energy converters (WECs) typically requires estimation of the wave excitation force (WEF). It was recently discovered that, if model-based WEF estimation is involved, the sensitivity of WEC control to model errors can change significantly, for both single WECs and WEC arrays. While previous results are numerical (MPC-based), this paper provides an analytical analysis of the coupling between WEC control and WEF estimation under model errors. The analysis is based on regular waves, unconstrained motion, and assumes noise-free, transient-free WEF estimation and ideal optimal velocity tracking. For the single-body case, errors in damping, mass, and stiffness are analysed. For the array case, errors in local models, i.e., ignorance of mutual radiation forces, are investigated. By straightforward derivation, insightful results are obtained, showing how model errors affect power capture differently with or without estimation. The analysis appears consistent with existing numerical results and generates guidelines for control system design of WECs.

Keywords: Sliding mode control, Linear robust control, Nonlinear robust control
Abstract: Floating Offshore Wind Turbines (FOWTs) face cost challenges driven by the large structural mass required for stability; as a result, the Levelized Cost of Energy (LCOE) of electricity from FOWTs can be high. To reduce foundation mass, this paper investigates active stabilization leveraging wave power in a Wave-Augmented Floating Offshore Wind Turbine (WAFOWT) device, in which three Oscillating Water Column (OWC) chambers embedded in the foundation provide control torque. By controlling the air mass flow through chamber valves, it is possible to create a differential torque about the platform pitch axis. A dynamic model is presented that couples the platform rigid-body dynamics with the chamber pressure dynamics. A Hamiltonian-based Sliding Mode Control (SMC-H) strategy is proposed in this work; this control approach uses the system's Hamiltonian as a sliding surface to achieve a balance between stability and optimality. Simulation results for a representative regular sea state demonstrate that the proposed SMC-H controller significantly reduces platform pitch motion compared to passive operation and linear-quadratic control benchmarks.

Keywords: Predictive control, Renewable Energy, Neural networks
Abstract: Model Predictive Control (MPC) can improve the energy capture of Wave Energy Converters (WECs), but it requires accurate future wave excitation force information. This paper proposes a direct excitation-force forecasting framework that combines a Long Short-Term Memory (LSTM) encoder with a Denoising Diffusion Probabilistic Model (DDPM) and integrates the predictor with MPC. Excitation-force labels are generated offline from wave elevation and the excitation impulse response function, allowing the model to map wave-elevation history directly to future excitation force and bypass online force estimation. The framework is trained under Pierson-Moskowitz (PM) irregular sea states and evaluated using WEC-Sim for a heaving point-absorber WEC. Relative to an autoregressive (AR) baseline, the selected 100-step, 5-sample LSTM-Diffusion predictor reduces PM mean and worst-case per-window RMSE by 24.34% and 39.62%, and increases generated energy by 19.3% over a 100 s control interval. When tested under a different JONSWAP (JS) irregular-wave condition, the PM-trained model still outperforms AR in both forecasting accuracy and generated energy, suggesting its robustness and ability to generalize across wave spectra.

Keywords: Control applications, Optimization, Renewable Energy
Abstract: Nonlinear dynamics and operational constraints pose significant challenges for the computational tractability of receding-horizon optimal control problems in wave energy systems. Direct transcription of optimal control problems yields nonlinear programming (NLP) formulations whose dimension and sparsity structure, two key drivers of computational cost, are strongly influenced by the chosen discretisation framework. This paper introduces a systematic approach for assessing discretisation frameworks prior to the formulation and solution of a full nonlinear receding-horizon wave energy converter (WEC) optimal control problem. The approach evaluates the trajectory-reconstruction accuracy and sparsity characteristics induced by compact- and global-support discretisation frameworks, relating these characteristics to the dimension and structure of the resulting NLP formulations. The results reveal a trade-off between approximation accuracy and sparsity preservation. While global-support schemes based on Half-range Chebyshev--Fourier and Legendre expansions remain attractive when their spectral-convergence properties can be exploited, Hermite--Simpson provides a particularly favourable balance between low-to-moderate-resolution approximation accuracy, NLP dimension, and sparsity, identifying it as a promising discretisation framework for nonlinear receding-horizon WEC optimal control.

Keywords: Predictive control, Linear systems
Abstract: This paper proposes a data-driven control method for preview tracking problems that aims to reduce tracking errors and output-constraint violations under disturbances and measurement noise while enforcing hard constraints on the input and its increment. The proposed method builds on Data-enabled Predictive Control (DeePC), decomposes the control input into tracking and regularized correction components, and reserves admissible input and input-increment ranges for the regularized correction component as a budget. The tracking component is optimized under constraints tightened according to this budget, and a two-stage optimization sequentially determines the tracking and regularized correction components. This construction ensures that, whenever both stage-wise optimization problems are feasible, the resulting composite input and its increment satisfy the original input and input-increment constraints. In the regularized correction stage, the regularized correction component and the associated data-driven trajectory representation are re-optimized within the reserved budget, allowing the reserved budget for the regularized correction component to be used in response to the updated input-output history under disturbances and measurement noise. Numerical simulations on a discrete-time double-integrator system show that, compared with standard DeePC, the proposed method reduces tracking errors and the number of steps with output-constraint violations from 15 to 0 under a step disturbance and from 20 to 0 under a step disturbance with measurement noise.

Keywords: Predictive control, Observers, Iterative learning control
Abstract: Control of nonlinear (NL) systems with high levels of uncertainty is practically relevant and theoretically challenging. This paper presents a numerical investigation of an adaptive NL model predictive control (MPC) technique that relies entirely on online system identification without prior modeling, training, or data collection. In particular, the paper extends predictive cost adaptive control (PCAC) for linear systems, which is an extension of generalized predictive control, to NL systems. NL PCAC (NPCAC) uses recursive least squares (RLS) with subspace of information forgetting (SIFt) to identify a discrete-time, pseudo-linear, input-output model, which is used with iterative MPC for NL receding-horizon optimization. The performance of NPCAC is illustrated using polynomial, Fourier, and cubic-spline basis functions.

Keywords: Predictive control, Optimization, Automotive applications
Abstract: We present laOPT, a native C++ toolbox for high-performance optimal control applications. The optimal control problem formulated by the user through an intuitive high-level interface is transcribed into a nonlinear program (NLP). laOPT offers solver interfaces to interior point and sequential quadratic programming (SQP) methods. In particular, the laOPT SQP solver can interface with various established QP solvers. Thanks to template programming, C++ compilers can optimize the solver executable for maximum runtime performance. We demonstrate the capabilities of laOPT on a high-speed mini race car experiment and benchmark against state-of-the-art optimization packages.

Keywords: Predictive control, Optimization, Energy Systems
Abstract: Self-powered control systems harvest disturbance energy to power the controller, eliminating reliance on external power sources. These systems are of interest in applications requiring autonomous operation without access to grid power or periodic recharging. Control inputs are constrained by finite energy storage and the requirement to prevent energy depletion during operation. This paper proposes a Model Predictive Control (MPC) approach that enforces these constraints while minimizing a quadratic performance objective. Nonconvex energy bounds are addressed using convex–concave approximations to enable efficient real-time optimization. However, nonlinear state evolution can render these approximations infeasible between MPC iterations. To address this, a feasibility restoration algorithm is introduced that minimally adjusts the convexification point while preserving real-time computational efficiency. The approach is demonstrated through simulation of a stochastically excited mechanical linkage system, showing that convex-concave MPC with feasibility restoration enables effective control of self-powered systems with guaranteed recursive feasibility.

Keywords: Predictive control, Optimization, Mining, minerals and petroleum
Abstract: This work demonstrates how model‑predictive control stabilized a highly variable ultra‑fines coal dewatering circuit, reducing moisture variability, energy intensity, and hydraulic disturbances while improving filter and thickener performance. Results highlight the value of model predictive control in complex solid-liquid separation systems.

Keywords: Predictive control, Power systems, Energy Systems
Abstract: Power systems are integrating more distributed energy resources (DERs) to meet decarbonization targets. Yet inverter-based generation reduces system inertia and increases the need for fast-acting dynamic ancillary services. Virtual power plants (VPPs) aggregate heterogeneous DERs to provide such services. However, existing approaches do not directly combine prescribed dynamic responses for fast frequency and voltage regulation with explicit enforcement of device- and distribution-network constraints in co-located VPPs. This paper proposes a model predictive control (MPC) framework for dynamic VPPs that tracks grid code-specified behaviour encoded by a desired transfer function while enforcing device- and feeder-level constraints. The framework implements a non-uniform prediction horizon that preserves fine near-term resolution for fast disturbance response while extending look-ahead without uniformly increasing computational burden. Case studies on a modified IEEE 33-bus feeder demonstrate close tracking of frequency and voltage regulation targets with practical real-time feasibility under suitable disturbances, and graceful degradation when requests exceed VPP capacity. The modular design accommodates diverse grid codes and DER portfolios, positioning the framework as a practical tool for evolving ancillary service markets.

Keywords: Automotive applications, Distributed control, Predictive control
Abstract: Autonomous vehicles have become a major drive for development in various fields, including control systems technology. In the context of driverless rovers, several available control solutions implement a form of decentralized control, where the interaction between lateral and longitudinal dynamics is neglected. In this paper we propose a novel distributed model predictive control strategy for trajectory tracking in the velocity space, that takes into consideration the interaction between dynamics while keeping the computational demands similar to decentralized controllers. We test our approach using IPGCarMaker for the car dynamics simulation, and show that it improves performance --measured by the cost of closed-loop trajectories-- by an average of 6% when compared to a decentralized control scheme.

Keywords: Automotive applications, Predictive control, Control Technology
Abstract: Torque Vectoring (TV) control is an effective strategy to improve the lateral performance of a vehicle with an additional yaw moment generated by unevenly distributing the torque demand among the individual wheels. However, dangerous total torque over-actuation may occur if the high-level controller requests yaw moments that do not consider individual wheel torque constraints and driver's longitudinal demand, particularly in powertrains limited to positive torques only. This paper proposes an MPC-based TV architecture in which admissible yaw moment bounds are computed online through a daisy-chain greedy allocation of the driver-requested total torque. The resulting bounds constrain the MPC optimization, ensuring that the downstream torque allocator can satisfy both wheel torque limits and the driver-imposed longitudinal demand. The approach is validated in a high-fidelity simulation environment, showing that the proposed strategy preserves the desired longitudinal behavior by only moderately worsening lateral performance.

Keywords: Automotive applications, Reinforcement learning, Robust control
Abstract: Automotive powertrain systems exhibit strong nonlinearities and parametric variations that limit the performance of controllers designed from nominal models. While deep reinforcement learning (DRL) can address such complexities, policies trained in simulation often lack robustness under real-world uncertainties, especially when extensive domain randomization is applied. This paper proposes a hybrid control framework that integrates model-based control with domain-randomized DRL for robust vibration suppression of nonlinear powertrain systems. The combined control structure is theoretically formulated within a latent Markov decision process, providing a unified framework for learning under randomized dynamics while preserving a model-based control backbone. A model-based H2 controller generates a baseline control action, while a learning-based policy compensates for nonlinear dynamics and parameter uncertainties. The learning-based component is optimized using Twin Delayed Deep Deterministic Policy Gradient (TD3) with recurrent neural networks, enabling stable policy learning under wide domain randomization. The proposed approach is validated through numerical simulations on a powertrain model with backlash nonlinearity. The results demonstrate improved learning stability and robust vibration suppression under significant parameter variations.

Keywords: Automotive applications, Simulation
Abstract: Vehicle stability and driving safety have been widely investigated in the context of in-wheel motor systems. This work proposes a torque assistance strategy based on a S-shaped sigmoid functions to improve lateral stability, while maintaining low computational demand. The first component of the torque assistance strategy compensates for lateral load imbalance by analyzing vertical tire forces in the z-axis direction. The second component employs a sigmoid function to avoid tire saturation according to an anticipative factor, reducing the motor torque before unstable or risky conditions. The analytical nature of the sigmoid function allows for a significant reduction in the computational burden while assisting vehicle stability through the active redistribution of wheel torques. Driver-in-the-loop simulations results demonstrate that the proposed strategy reduces the understeer gradient by up to 25%, maintains the yaw rate within safety limits, and decreases the required driver input on the accelerator and brake pedals. These results confirm that the proposed approach improves vehicle behavior in cornering maneuvers by preventing tire saturation in advance, without imposing a significant computational burden and allowing easy integration into real-time embedded control systems.

Keywords: Automotive applications, Transportation systems, Control applications
Abstract: This paper addresses the problem of energy-efficient thermal management in Battery Electric Vehicles (BEVs), with the goal of extending driving range while ensuring passenger comfort and battery thermal safety. A hierarchical Nonlinear Model Predictive Control (NMPC) framework is proposed, in which the supervisory layer explicitly balances energy consumption and cabin temperature reference tracking over a long horizon and translates this trade-off into optimal actuator command sequences for the integrated heating, ventilation, and air conditioning (HVAC) system and the battery thermal management (BTM) system. These command sequences are then enforced by a lower-layer NMPC, which introduces local corrections to compensate disturbances and modeling uncertainty while tracking an externally specified cabin temperature reference trajectory. To enable predictive control of the highly nonlinear and strongly coupled thermal dynamics, data-driven neural network models are employed as prediction models within both control layers. The proposed approach is developed and validated using a high-fidelity BEV thermal simulator developed at Politecnico di Torino in collaboration with Stellantis N.V., and benchmarked against an industrial rule-based (RB) control strategy over the WLTC driving cycle. Simulation results demonstrate a significant reduction in energy consumption compared to the baseline strategy, while satisfying comfort and battery temperature constraints.

Keywords: Automotive applications, Energy Storage, Reinforcement learning
Abstract: Cell-to-cell imbalance in automotive battery packs reduces usable energy, accelerates degradation, and constrains operational safety. Conventional passive cell balancing relies on fixed thresholds and rule-based logic that are difficult to tune and lack adaptability under cell variability and changing operating conditions. This paper presents a reinforcement learning (RL)-based passive cell balancing controller for a production-representative automotive battery pack. The controller is implemented in a high-fidelity MATLAB/Simulink simulation environment combining an equivalent circuit battery model with a module-level thermal model. A soft actor–critic agent learns continuous per-cell balancing duty commands to reduce relative charge imbalance while respecting voltage safety constraints and minimizing unnecessary balancing activity. Simulation results show rapid imbalance reduction during early stages and a natural tapering of balancing effort near convergence, with all cells maintained within safe voltage limits. These results demonstrate the potential of RL to improve the adaptability and performance of passive cell balancing compared to conventional rule-based strategies

Keywords: Adaptive control, Linear parameter-varying systems, Gain scheduling
Abstract: This paper presents an efficient kernel design for Bayesian Optimization (BO) in a Linear Parameter-Varying (LPV) framework. Traditional controller tuning methods often struggle with time-varying plant dynamics, leading to suboptimal performance. By leveraging a problem-specific kernel within BO, the proposed approach aims to enhance the accuracy and efficiency of tuning procedures using BO under the assumption that the plant dynamics are approximately known. The method is validated on a representative case study, demonstrating that incorporating prior model knowledge results in improved convergence speed and better estimation of optimal tuning parameters compared to standard kernels such as the Mat'{e}rn 5/2 kernel, particularly when only a very limited number of samples is available, thereby reducing the need for manual adjustments and enhancing overall system performance.

Keywords: Adaptive control, Predictive control, Control applications
Abstract: This paper addresses position control of an electrostatic levitation system under terrestrial gravity. Conventional PID controllers often fail to maintain stability under abrupt variations in the levitated sample's charge. To address this issue, we propose a generalized predictive control (GPC) scheme for electrostatic levitation. A data-driven model is identified online using recursive least squares (RLS). The resulting GPC controller stabilizes the levitated position despite rapid charge changes. Compared with a baseline PID controller, the proposed method yields smoother responses, improved stability, and stronger disturbance rejection.

Keywords: Renewable Energy, Adaptive control, Gain scheduling
Abstract: This paper proposes an adaptive control approach for floating offshore wind turbines (FOWTs) subjected to time-varying frequency waves. The proposed approach switches controllers based on the estimated incident wave frequency. These selectable controllers consist of pre-designed H-infinity controllers and their interpolated variants. While this work is fundamentally based on a nonlinear FOWT model, the H-infinity controllers are designed and analyzed using an approximately linearized model. System analyses focus on stability and the internal model principle for the wave-frequency mode, which directly governs control performance. The analysis demonstrates that each H-infinity controller, including the interpolated ones, adequately incorporates the wave-frequency mode. Furthermore, each wave-frequency-frozen linearized closed-loop system is shown to be asymptotically stable across the considered frequency range as a necessary condition. To estimate the incident wave frequency, a zero up-crossing method enhanced by a low-pass filter is employed. Simulation results validate that the proposed system achieves successful control performance against time-varying frequency waves, even under simultaneous and abrupt wind speed changes. Consequently, the efficacy of the proposed adaptive control system is confirmed.

Keywords: Adaptive control, Learning, Mobile Robots
Abstract: This paper presents the FL-DeePO framework for balancing an autonomous bicycle. It combines an inner-loop Feedback Linearization (FL) with an outer-loop based on the Data-Enabled Policy Optimization (DeePO) algorithm. FL stabilizes the nonlinear, open-loop unstable bicycle and partially cancels its nonlinearities, while DeePO adapts online to unmodeled dynamics and time-varying disturbances. Validated in high-fidelity simulation and on an instrumented bicycle, FL-DeePO outperforms FL-only and deep RL baselines in lean angle and lean rate tracking.

Keywords: Sliding mode control, Adaptive control, Cooperative control
Abstract: This paper presents a resilient cooperative adaptive cruise control (CACC) framework for a platoon of connected and autonomous vehicles (CAVs) in the presence of false data injection (FDI) attacks, actuator faults, and external disturbances. To mitigate these adverse effects, an observer-based control framework is proposed. The design incorporates a third-order sliding mode observer to estimate the lumped uncertainty arising from actuator faults and external disturbances, as well as an extended state observer to estimate FDI attacks and disturbances entering the error dynamics. The resulting estimates are incorporated into a compensation-based nonlinear control law to actively counteract FDI attacks and actuator faults in real time. Lyapunov-based stability analysis is provided, and MATLAB/Simulink simulation results demonstrate that the proposed approach maintains safe inter-vehicle spacing and stable platoon behavior under adverse operating conditions.

Keywords: Control applications, Control Technology, Machine learning
Abstract: Industrial feed drives are primarily controlled using a cascaded control structure, which largely governs the dynamic behavior of machine tools. Velocity feedforward controllers can enhance the feed drive performance. However, conventional designs offer limited accuracy, while model-inversion-based approaches, though more precise, are sensitive to unmodeled disturbances and nonlinearities that make tuning challenging. This work proposes an analytical flat model for an inversion-based feedforward controller, whose tuning is shifted to Bayesian optimization as a data-driven framework to account for these unmodeled effects. The controller is tuned directly in the time domain, minimizing the tracking error in single-axis operation and the contour error in multi-axis operation. Experimental validation on a five-axis milling machine confirms the effectiveness of the proposed approach, with performance improvements exceeding 59% over the classical cascaded control structure and 25% over an exact inverse feedforward controller.

Keywords: Control applications, Mechanical systems, Control architectures
Abstract: This study proposes a hierarchical control architecture composed of a Model-Based Controller (MBC) and a Model Error Compensator (MEC) for a hydraulic system with one pump and two actuators, and experimentally validates its effectiveness using an actual machine. In conventional hydraulic systems, simultaneous actuation tends to cause flow to concentrate on the actuator with the lower load, which may prevent the system from achieving the desired flow characteristics depending on the design. To address this issue, the proposed method assigns independent MECs to each actuator and generates per-actuator flow commands through the MBC, enabling effective compensation for the flow imbalance. Experimental comparisons with and without the compensation structure confirmed that the proposed system provides notable improvements in overshoot suppression, response delay reduction, and mitigation of hydraulic interference among actuators.

Keywords: Control Technology, Mechatronic systems, Control applications
Abstract: Designing vibration-suppression feedforward (FF) compensation is crucial to achieve fast-response and high-accuracy control performance in precision servo mechanisms with high-order resonant modes. However, traditional model-based FF design approaches often require substantial time and effort to determine a plant model for the FF compensation design. In this study, we propose a performance-oriented model-based FF (PO-MBFF) compensation design method that utilizes data-driven response prediction. The proposed PO-MBFF method is used to design FF compensators within the framework of existing model-based approaches, and the plant model used for the FF design is automatically determined through predicted response-based performance optimization. The effectiveness of the PO-MBFF method in terms of control performance and design efficiency was demonstrated through numerical simulations using a piezoelectric-actuated fast-steering mirror.

Keywords: Mechanical systems, Linear parameter-varying systems, Estimation
Abstract: Industrial lifting devices enable fast and efficient logistics in warehouses all over the world. As storage space is expensive, such vehicles are built to reach racks at enormous heights, which makes them vulnerable to tipping over. To ensure tip-over safety, they rely on robust measurement of the current lift height. In contrast, common sensors such as optical systems or cable-pull sensor are often unreliable or expensive to protect against environmental influences in harsh working environments. In this paper, we present three different online approaches applicable in the industrial context using a low-cost and often already present inertial measurement unit (IMU) at the tip of the mast. The first method reconstructs the IMU orientation with a complementary filter and integrates the vertical acceleration, the second method uses the alpha sliding window infinite Fourier Transform (aSWIFT) algorithm for real-time frequency estimation and relates it to a frequency-lift-height model and the third method uses an extended Kalman filter (EKF) and directly estimates the parameter. All three methods show promising results and can be applied in an industrial context, depending on the application scenario.

Keywords: Mechatronic systems, Control architectures, Nonlinear systems
Abstract: Piezoelectric nanopositioning systems are often limited by lightly damped structural resonances and the gain--phase constraints of linear feedback, which restrict achievable bandwidth and tracking performance. This paper presents a dual-loop architecture that combines an inner-loop non-minimum-phase resonant controller (NRC) for active damping with an outer-loop tracking controller augmented by a constant-gain, lead-in-phase (CgLp) reset element to provide phase lead at the targeted crossover without increasing loop gain. We show that aggressively tuned CgLp designs with larger phase lead can introduce pronounced higher-order harmonics, degrading error sensitivity in specific frequency bands and causing multiple-reset behavior. To address this, a shaping filter is introduced in the reset-trigger path to regulate the reset action and suppress harmonic-induced effects while preserving the desired crossover-phase recovery. The proposed controllers are implemented in real time on an industrial piezo nanopositioner, demonstrating an experimental open-loop crossover increase of approximately 55~Hz and a closed-loop bandwidth improvement of about 34~Hz relative to a well-tuned linear baseline.

Keywords: Predictive control, Process control, Linear systems
Abstract: This paper presents two explicit Model Predictive Control (MPC) formulations for linear systems parameterized in terms of design variables. Such parameter-dependent behavior commonly arises from operating-point–dependent linearization of nonlinear systems as well as from variations in mechanical, electrical, or thermal properties associated with the design of the process or system components. In contrast to explicit MPC (eMPC) approaches that treat design parameter dependencies as disturbances, the proposed methods incorporate the parameters directly into the system matrices in an affine manner. However, explicitly incorporating these dependencies significantly increases the complexity of eMPC formulations due to the resulting nonlinear terms involving decision variables and parameters. We address this complexity by proposing two approximation methods. Both methods are applied on two examples and their performances are compared with respect to the exact eMPC implementation.

Keywords: Predictive control, Real-time systems, Linear systems
Abstract: Recent cloud and edge control platforms necessitate sophisticated resource management to satisfy real-time requirements of multiple co-located tasks. Existing dynamic resource allocation strategies typically adjust resources reactively, based on the current criticality of the control tasks. This can lead to situations hard to recover from with a late intervention, especially if computational resources are scarce. To address this limitation, we propose a predictive resource allocation framework, leveraging a Model Predictive Control (MPC) strategy with a variable update rate. We model the evolution of uncertainty around the predicted trajectories as a function of the update rate, and use it to make proactive decisions on the update rate of the controllers. The approach is demonstrated via simulation on a relevant use-case, improving overall performance and resource utilization compared to a traditional dynamic allocation approach.

Keywords: Predictive control, Reinforcement learning, Control architectures
Abstract: Liquid-level control in coupled tank systems poses challenges due to their nonlinearity and multiple time scales. Solving this problem is crucial for quality control, production optimization, and process flexibility in industries such as oil refining, water treatment, and chemical mixing. The use of coupled tanks is an effective way to simplify a multi-time system, as it captures both short- and long-term impacts within this nonlinear framework. As a solution, the model predictive control (MPC) offers a viable approach for short-horizon control, while model-free reinforcement learning (RL) methods have proven effective for addressing long-term effects in control systems. Two approaches to integrate MPC and RL, leveraging their advantages for short-term path tracking while minimizing long-term effects in coupled tank systems, are presented. Experimental findings indicate that the first method, adjusting the MPC setpoint, is beneficial in scenarios where long-term minimization is crucial, reducing the tank 2 level by 10.4% compared to short-horizon MPC. The second method, which integrates an RL control action into the MPC output, is advantageous when this new sudden change in the output is feasible for the actuator. With 3.5 times greater maximum control effort variation, steady-state error is reduced by 61.3% compared to short-horizon MPC. Both methods effectively address unmodeled dynamics and reduce the final steady-state error.

Keywords: Predictive control, Sensor fusion, Robust control
Abstract: This paper proposes a Robust Subsystem-based Impedance Model Predictive Control (RSI-MPC) framework for the task-space control of flexible manipulators. Controlling flexible-link manipulators poses significant challenges due to their distributed dynamics, structural vibrations, and sensitivity to external disturbances. To address these issues, the flexible structure is modeled kinematically and dynamically as an open chain of rigid bodies. This formulation uses feedback from a network of Inertial Measurement Unit (IMU) sensors, enabling accurate state estimation while maintaining compatibility with real-time implementation in closed-loop configurations. Building on the approximated dynamic model of the open-chain system, the proposed RSI-MPC method ensures robust impedance behavior under model uncertainties and varying interaction conditions. Experimental results demonstrate that the RSI-MPC significantly improves the precision of interaction force handling while simultaneously generating smoother and lower-magnitude control inputs, validating its effectiveness in real-time manipulation tasks.

Keywords: Predictive control, Simulation, Control applications
Abstract: We study finite-horizon budget allocation as a closed-loop economic control problem and evaluate receding-horizon Model Predictive Control (MPC) relative to reactive budgeting policies. Budgets are allocated periodically under execution noise and operational constraints, while return efficiency may evolve over time. Using a controlled simulation framework motivated by digital marketing, we compare reactive pacing to MPC across environments with increasing degrees of non-stationarity. Our results show that non-stationarity alone does not justify predictive control. When return dynamics are stationary or evolve through unpredictable stochastic drift, MPC offers no systematic advantage over reactive baselines. By contrast, when return efficiency exhibits predictable structure over the planning horizon, that is captured through an underlying model, MPC consistently outperforms reactive budgeting by exploiting intertemporal trade-offs.

Keywords: Predictive control, Machine learning, Modeling
Abstract: This abstract presents the results of our IEEE TCST paper [1]. Modern industrial processes, including carbon capture systems, energy systems, and large-scale chemical plants, often exhibit high-dimensional nonlinear dynamics, strong coupling, and strict operational constraints, which make safe, efficient, and economically optimal operation challenging. Model predictive control (MPC) is well suited to such systems because it can explicitly handle constraints and optimize performance. However, for complex nonlinear processes, deriving such a model from first principles is often difficult.

In our paper [1], we proposed a deep neural Koopman operator (DNKO)-based economic model predictive control (EMPC) framework for shipboard post-combustion carbon capture systems. The proposed DNKO model uses neural-network-based lifting and recurrent structures to learn latent representations from partial state measurements and historical trajectories, enabling accurate prediction of key system outputs and economic cost despite limited measurements and time-varying operating conditions. Based on the learned Koopman representation, a constrained EMPC problem is formulated, where the linear evolution of lifted states enables a convex optimization problem suitable for real-time implementation. The proposed method explicitly accounts for hard output constraints while optimizing economic performance and carbon capture rate. Extensive simulations under four representative ship operating conditions demonstrate that the approach achieves improved economic performance and higher carbon capture efficiency while maintaining safe operation [1].

Reference [1] M. Han and X. Yin. Deep neural Koopman operator-based economic model predictive control of shipboard carbon capture system. IEEE Transactions on Control Systems Technology, 33(6):2064-2079, 2025

Keywords:
Abstract: Power systems around the world are being modernized to address environmental concerns, reduce costs, and guarantee access to electricity all the time. Four main criteria for this upgrade are efficiency, reliability, resiliency and sustainability. Recent advances in various technologies are the key enablers for this modernization. Nevertheless, such physical systems are becoming overwhelmingly large scale and stochastic with highly complex dynamics, coupled with millions of human interactions. The design and operation of these systems needs major innovations in computational techniques. In this talk, we first discuss some major challenges behind the modernization of power grids and explain why addressing them involves many different fields.

Then, we focus on three topics of optimization, learning, and control for power systems, which all need major revolutions in computational techniques. We study how recent advances in AI and machine learning can assist with addressing some of these challenges. We offer case studies on the grids for California, Texas, and different parts of Europe.