Neural network controllers trained via behavior cloning are increasingly deployed in cyber-physical systems (CPS), yet practitioners lack tools to trace controller failures back to training data. Existing data attribution methods assume i.i.d. data and standard loss targets, ignoring CPS-specific properties: closed-loop dynamics, safety constraints, and temporal trajectory structure. We propose IF-CPS, a modular influence function framework with three CPS-adapted variants: safety influence (attributing constraint violations), trajectory influence (temporal discounting over trajectories), and propagated influence (tracing effects through plant dynamics). We evaluate IF-CPS on six benchmarks across diagnosis, curation, and safety attribution tasks. IF-CPS improves over standard influence functions in the majority of settings, achieving AUROC 1.00 in Pendulum (5-10% poisoning), 0.92 vs. 0.50 in HVAC (10%), and the strongest constraint-boundary correlation (Spearman ρ= 0.55 in Pendulum).

Model-based controllers learned from data have the biases and noise of their training trajectories, making it important to know which trajectories help or hurt closed-loop performance. Influence functions, widely used in machine learning for data attribution, approximate this effect through first-order parameter-shift surrogates, avoiding costly retraining. Applying them to stochastic LQR, however, is nontrivial because the cost depends on the learned dynamics through the Riccati equation, and the process-noise covariance is estimated from the same residuals. We develop a three-level influence hierarchy that accounts for both channels.

This paper presents a hierarchical two-stage framework for multi-robot task allocation and trajectory optimization in asymmetric task spaces: (1) a sequential auction allocates tasks using closed-form bid functions, and (2) each robot independently solves an optimal control problem for energy-minimal trajectories with a physics-based battery model, followed by a collision avoidance refinement step using pairwise proximity penalties. Event-triggered warm-start rescheduling with bounded trigger frequency handles robot faults, priority arrivals, and energy deviations. Across 505 scenarios with 2-20 robots and up to 100 tasks on three factory layouts, both energy- and distance-based auction variants achieve 11.8

ECG digitization could unlock billions of archived clinical records, yet existing methods collapse on real-world images despite strong benchmark numbers. We introduce VLM-in-the-Loop, a plug-in quality assurance module that wraps any digitization backend with closed-loop VLM feedback via a standardized interface, requiring no modification to the underlying digitizer. The core mechanism is tool grounding: anchoring VLM assessment in quantitative evidence from domain-specific signal analysis tools. In a controlled ablation on 200 records with paired ground truth, tool grounding raises verdict consistency from 71% to 89% and doubles fidelity separation (ΔPCC 0.03 → 0.08), with the effect replicating across three VLMs (Claude Opus 4, GPT-4o, Gemini 2.5 Pro), confirming a pattern-level rather than model-specific gain. Deployed across four backends, the module improves every one: 29.4% of borderline leads improved on our pipeline; 41.2% of failed limb leads recovered on ECG-Digitiser; valid leads per image doubled on Open-ECG-Digitizer (2.5 → 5.8). On 428 real clinical HCM images, the integrated system reaches 98.0% Excellent quality. Both the plug-in architecture and tool-grounding mechanism are domain-parametric, suggesting broader applicability wherever quality criteria are objectively measurable.

Data-Enabled Predictive Control (DeePC) has emerged as a powerful framework for controlling unknown systems directly from input-output data. For nonlinear systems, recent work has proposed selecting relevant subsets of data columns based on geometric proximity to the current operating point. However, such proximity-based selection ignores the control objective: different reference trajectories may benefit from different data even at the same operating point. In this paper, we propose a datamodel-based approach that learns a context-dependent influence function mapping the current initial trajectory and reference trajectory to column importance scores. Adapting the linear datamodel framework from machine learning, we model closed-loop cost as a linear function of column inclusion indicators, with coefficients that depend on the control context. Training on closed-loop simulations, our method captures which data columns actually improve tracking performance for specific control tasks. Experimental results demonstrate that task-aware selection substantially outperforms geometry-based heuristics, particularly when using small data subsets.

For efficient roll-to-roll (R2R) production of flexible electronic components, a precise R2R transfer peeling process is essential, requiring accurate modeling and control. This paper introduces a novel approach to confining the dynamics of a nonlinear R2R mechanical peeling system within a convex set known as a norm-bounded linear differential inclusion (NLDI). This method utilizes constraints on uncertain system variables to create a tighter NLDI representation compared to other convexification techniques. Moreover, it offers drastically reduced computational cost compared to previous methods applied to convexify the R2R peeling system. The NLDI is employed to generate an H infinity-optimal controller for the R2R peeling system, and both simulations and experiments demonstrate better dynamic performance compared to other controllers for R2R transfer.

This paper presents a Model-Inspired Distributionally Robust Data-enabled Predictive Control (MDR-DeePC) framework for systems with partially known and uncertain dynamics. The proposed method integrates model-based equality constraints for known dynamics with a Hankel matrix-based representation of unknown dynamics. A distributionally robust optimization problem is formulated to account for parametric uncertainty and stochastic disturbances. Simulation results on a triple-mass-spring-damper system demonstrate improved disturbance rejection, reduced output oscillations, and lower control cost compared to standard DeePC. The results validate the robustness and effectiveness of MDR-DeePC, with potential for real-time implementation pending further benchmarking.

When a controller is designed from an identified model, its performance ultimately depends on the trajectories used for identification, but 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 LQR pipeline. On the identification side, the quadratic loss admits an exact leave one trajectory out parameter shift and a reusable first order approximation with a Neumann series error bound. On the control side, we implicitly differentiate through the 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 and speedups of 7 to 60 times on linear systems of dimension 2 to 10.

We extend the Datamodels framework from supervised learning to Model Predictive Path Integral (MPPI) control. Whereas Datamodels estimate sample influence via regression on a fixed dataset, we instead learn to predict influence directly from sample cost features, enabling real-time estimation for newly generated samples without online regression. Our influence predictor is trained offline using influence coefficients computed via the Datamodel framework across diverse MPPI instances, and is then deployed online for efficient sample pruning and adaptive constraint handling. A single learned model simultaneously addresses efficiency and safety: low-influence samples are pruned to reduce computational cost, while monitoring the influence of constraint-violating samples enables adaptive penalty tuning. Experiments on path-tracking with obstacle avoidance demonstrate up to a 5× reduction in the number of samples while maintaining control performance and improving constraint satisfaction.

Autonomous Mobile Robots (AMRs) play a crucial role in transporting materials across expansive manufacturing facilities and warehouses. Their successful deployment relies on three major factors: task allocation, task scheduling, and trajectory planning. These processes collectively shape the efficiency and effectiveness of AMRs in complex manufacturing and warehouse environments. This study focuses on AMR trajectory planning, emphasizing energy efficiency beyond traditional methods. We present a physics-oriented AMR model and an optimal control strategy to generate energy-optimized routes. Through simulation studies across different scenarios, we evaluate the efficacy of diverse numerical solutions and compare two different AMR designs, one with Ackermann steering and the other with Mecanum steering. Our results indicate that the proposed approach yields a 5-10% energy advantage over traditional shortest-path algorithms, without compromising computational integrity and efficiency. The saving is more pronounced for the AMR with Ackermann steering. These findings are also validated with an experimental study.

Abstract For efficient roll-to-roll (R2R) production of flexible electronic components, a precise R2R transfer peeling process is essential, requiring accurate modeling and control. This paper introduces a novel approach to confining the dynamics of a nonlinear R2R mechanical peeling system within a convex set known as a norm-bounded linear differential inclusion (NLDI). This method utilizes constraints on uncertain system variables to create a tighter NLDI representation compared to other convexification techniques. Moreover, it offers drastically reduced computational cost compared to previous methods applied to convexify the R2R peeling system. The NLDI is employed to generate an H_8-optimal controller for the R2R peeling system, and both simulations and experiments demonstrate better dynamic performance compared to other controllers for R2R transfer.

Dry transfer using Roll-to-Roll (R2R) mechanical peeling could significantly increase the throughput and efficiency of the production of 2D materials such as graphene and flexible electronics. Currently, such a R2R process does not exist in industry. For this dry transfer R2R process to be practical for industrial applications, the peeling angle between the growth substrate and the functional material needs to be precisely controlled. In this paper, a nonlinear state space representation of the R2R dry peeling process is formulated with the peeling front velocity variation as a disturbance input. This state space model is used to construct a linear parameter varying (LPV) representation of the system, and a methodology on how to bound the LPV representation within a convex polytopic linear differential inclusion (PLDI) set is presented. This PLDI representation is then used in a linear matrix inequality (LMI) optimization framework to design a full state feedback controller that minimizes the H∞ gain of the connection between the adhesion energy variation and the peeling front geometry. Simulation results demonstrate that this controller improves the precision of the R2R peeling angle, and this increase in precision enables higher web speed. Thus, this technique can be an enabling tool for making R2R mechanical peeling dry transfer of 2D materials a reality in industrial settings.

Non-aqueous drilling fluids are generally favored in risered deepwater drilling operations for their capability to significantly improve drilling performance. However, their use may increase the difficulty of gas kick detection and handling, due to the high solubility of gas at downhole pressure and temperature conditions. A novel managed pressure drilling (MPD) control strategy is proposed here to handle gas kicks and prevent riser gas unloading (RGU) events in cases of gas influxes in non-aqueous drilling fluids. With the proposed strategy, the dissolved gas kick is diluted to a less severe level after it passes the subsea blow-out preventers by injecting mud into the riser through the booster line. Since the dissolved gas concentration is decreased after dilution, it is easier to mitigate or even fully eliminate an RGU event by applying an appropriate MPD backpressure. This MPD backpressure keeps the kick in solution when it is in the riser, and only allows it to come out of solution after passing the MPD choke. The proposed strategy's effectiveness is demonstrated using multiphase flow model simulations. The modeling approach considers the transient behavior of the two-phase gas-fluid mixture, and is validated against full-scale experimental data. Simulation results for a typical deepwater drilling operation show that the proposed MPD control strategy with riser dilution can greatly improve both the safety and speed of handling gas kicks in deepwater drilling. The improved safety derives from the ability to eliminate RGU events, and the increased speed comes from handling the kick directly in the riser using MPD backpressure and booster line dilution rather than through the lengthy conventional process of circulating it out through the choke line. Sensitivity analysis is conducted to investigate how the performance of the proposed control strategy is affected by influx gas composition, reservoir depth, seawater column height, kick size, and dilution ratio. A main conclusion is that the backpressures required after dilution are within the operating limits of regular marine risers, and therefore the majority of gas influx cases can be handled without the need for high-pressure risers. This allows for the relatively straightforward deployment of the proposed control strategy in the field.

Roll-to-Roll (R2R) mechanical peeling for dry transfer has the potential to enable high throughput production of 2D materials and flexible electronics, while minimizing the environmental impact. The dry transfer process is highly nonlinear, involving challenging peeling front dynamics. In this study, a novel convexification scheme is developed based on polytopic linear differential inclusions (PLDIs). An LQR controller is used to establish that the performance of a linear approximation built using this PLDI-based method is consistently superior to that of a Taylor’s expansion-based approximation, demonstrating that the PLDI-based scheme is an effective control tool for complex systems, including the R2R mechanical peeling process.

With the development of factory automation and intelligent manufacturing system technology, an autonomous mobile robot (AMR) system has become an essential part of controlling the logistics management system within a facility. Relevant research about AMR path planning usually focuses on the fully autonomous environment that does not consider the uncertainty of human behavior. The behavior of human operators is unpredictable and therefore difficult to be integrated into the AMR system path planning analysis. In this paper, we propose an optimization algorithm for improving the pre-planned path considering the uncertainty of human behavior. Conditional value-at-risk constraints and chance constraints are considered in the optimization algorithm as the risk measurement to guarantee safety of operation. The performance of our approach is demonstrated through a 2-D AMR simulation, and the comparison of these two different risk measurements and their performance is also discussed.

With the development of factory automation, intelligent manufacturing system technology, and three-dimensional automated warehouses, an automatic ground vehicle (AGV) became an essential part of controlling the discrete logistics management system within a facility. The scope of the AGV application and technical capability have been rapidly developed in recent years. However, it is highly time-consuming and resource-intensive to develop a comprehensive AGV platform in both industrial setting and academic environment to design and control of an AGV system. This paper introduces a platform for conducting AGV research and deployment, which consists of the hardware prototyping and entire software system development. By using this platform, users can readily develop customized AGV systems or verify their self-developed algorithms.

Autonomous Mobile Robots (AMRs) have become extremely popular in the manufacturing domain, especially for processes involving large factory floors where these robots are used for transporting materials from one location to another. In an environment where there are multiple prioritized tasks to be completed by a school of AMRs, the overall planning problem can be broken down into three sequential steps: task allocation for the school of AMRs, task scheduling for each AMR, and trajectory planning for each individual AMR. This paper focuses on the trajectory generation procedure for each AMR. Unlike traditional approaches that only consider the location an AMR has to travel to during path planning, here, energy efficiency of the AMR is also considered. We present the physics-based model of the AMR as well as an optimal control formulation for energy-conscientious trajectory generation for the AMR. Methods to numerically solve this problem are discussed, and results are presented for each proposed algorithm on approximately 100 test cases, comparing both performance and computational efficiency. The results show that the presented energy-conscientious methods perform better in terms of energy usage (5-10%) compared to commonly-used shortest path techniques while maintaining similar computational and operational efficiency.

The use of Autonomous Mobile Robots (AMRs) for fast and efficient manufacturing has attracted the interest of academia and industry in recent times, especially due to significant improvements in computational efficiency. However, one of the biggest challenges in terms of controls is the optimal task assignment and scheduling of AMRs in order to finish the assigned tasks as quickly as possible, taking into account the priority of the tasks. This paper focuses on the single-AMR scheduling problem, i.e., once each AMR is assigned a set of tasks, the objective is to order these tasks efficiently while considering task priorities. The need for real-time algorithms to solve this problem renders exhaustive search algorithms inappropriate, since their focus is on the accuracy of the solution without considering time constraints. This paper proposes a gradient-based real-time approach for the scheduling problem based on a mathematical formulation in the structure of a regularized quadratic program. This scheduling algorithm is shown to perform better than a simulated annealing based pairwise exchange technique, which is a commonly used heuristic method, in terms of a defined cost metric. Therefore, the proposed algorithm allows for the generation of efficient realtime solutions to the scheduling problem for a single AMR.

AbstractChronic bradycardia, or slowing of heart rate, is common in preterm infants, and may often lead to neuropsychiatric disorders, developmental problems, and impaired cognitive functions in the long term. Therefore, early detection and treatment of bradycardia is important. To this end, we present a system identification-based approach to the prediction of bradycardia in preterm infants. This algorithm is based on the notion that the cardiovascular system can be treated as a dynamic system, and that under bradycardia, this system reacts abnormally due to temporal and spatial destabilization. This paper presents a proof-of-concept of the proposed methodology by testing its performance using electrocardiogram (ECG) data collected from ten preterm infants. We show that the proposed algorithm is correctly able to predict bradycardia occurrences (mean area under the receiver operating characteristic (ROC) curve = 0.782 and variance = 0.0039) while minimizing the training or burn-in period. The physical interpretation of the results using the system dynamics approach is discussed. The developed algorithm performs well on not only classifying normal to abnormal conditions, but also showing a trend of transition between the two conditions. Future work is also discussed to further improve the algorithm and implement the algorithm in the neonatal intensive care unit. Our proposed method is able to predict bradycardia using only ECG data with minimal training period and can be integrated into an automated system for bradycardia detection and treatment, and therefore, reduce the risks related to bradycardia in preterm infants.