In the field of human–robot interaction (HRI), achieving flexibility in human-accompanying within real-world environments holds great potential for various applications but also poses significant challenges. Traditional methods typically restrict robots to fixed positions relative to humans, such as tracking from behind, in front, or side-by-side, which limits robot adaptability in dynamic workspaces. This study introduces a novel human–companioning strategy that uses reinforcement learning (RL) to enable mobile robots to dynamically adjust their tracking positions according to varying conditions. An interaction space is defined to capture the relationship between the human and the robot while considering the environment, which serves as the basis for the state spaces in the DRL to assist the robot in adapting to environmental changes. A human–robot companion controller is developed by integrating model predictive path integral (MPPI) control with control barrier functions (CBFs), ensuring that the robot accurately follows the target’s movement in both position and orientation while avoiding obstacles and enhancing social acceptance and safety. The proposed approach is evaluated in real-world scenarios, both indoors and outdoors, and compared with those of other studies. The results show that the proposed method improves the success rate (SR) and tracking accuracy by at least 24% and 47%, respectively, while enhancing human comfort. Experiments demonstrate the robot’s ability to flexibly accompany a person walking at speeds of up to 1.7 m/s, dynamically adjusting its strategy without being confined to a fixed position. In addition, the robot respects the human’s intimate space to ensure safety, comfort, and effective obstacle avoidance.
Malicious cyber intrusions have posed an escalating threat to the operational reliability of transportation cyber–physical systems (TCPSs). Motivated by this issue, this study investigates the resilient security protection problem for active vehicle suspension systems (AVSSs) embedded within the TCPS framework against cyberattacks. First, an interval type-2 (IT-2) fuzzy model is constructed to tackle the uncertainties arising from time-varying spring stiffness and damping coefficients. Second, a secure-oriented multiinstant gain-scheduling (SMG) defense strategy is developed. By systematically utilizing normalized fuzzy weighted membership degrees (NFWMDs) over consecutive sampling instants, the proposed high-order strategy can not only alleviate conservatism in the stability criteria but also guarantee smooth vehicle operation in the presence of concurrent cyberattacks and road disturbances. Third, a dedicated slack variable technique tailored to the proposed SMG strategy is introduced to exploit the inherent algebraic properties of the NFWMDs, which yields more relaxed exponential stability conditions for AVSSs under stochastically triggered cyberattacks compared to those reported in recent studies. Finally, the hardware-in-the-loop (HIL) testing is performed to validate the superiority of the developed method.
This article focuses on the robust stabilization of large-scale Boolean networks (BNs) with a product-type bridging fault. First, the concept of product-type bridging fault is proposed on the basis of BNs with the minimal disjunctive normal form. Second, with the assistance of the adjacency matrix and index sets constructed through the ideal fixed point, several new sufficient conditions are proposed for the robust stabilization of large-scale BNs via the original constant controller. Third, when the robust stabilization cannot be achieved, an algorithm is given to improve the constant controller by isolating the faulty node. Finally, two examples are provided to illustrate the effectiveness of the results.
A model-free robust fuzzy tracking strategy for autonomous vehicles (AVs) is developed by integrating hybrid reinforcement learning (RL) with a zero-sum game framework. First, the vehicle dynamics are modeled by the Takagi–Sugeno fuzzy technique while simultaneously accounting for velocity variations and external disturbances. The tracking objective is then formulated as a two-player zero-sum game, in which the control input and the worst case disturbance are treated as two adversarial players, yielding a set of fuzzy game algebraic Riccati equations (FGAREs) whose analytical solutions are generally difficult to obtain. To address this issue, an offline parallel hybrid iteration (PHI) algorithm that combines value and policy iterations (PIs) is proposed to solve the FGAREs without requiring an initial stabilizing policy while achieving faster convergence. When the vehicle model is unavailable, an online model-free PHI scheme driven by Q-learning (QL) is further designed to update the fuzzy value function and the control policy using only measured data. Finally, simulation results are presented to demonstrate the effectiveness of the proposed method.
This article proposes an innovative prescribed performance control (PPC) scheme for nonlinear systems with abrupt load variations, featuring a dual-regulation mechanism with passive and active modes. In passive mode, the controller evaluates both the tracking error and its distance to the prescribed boundaries, enabling adaptive relaxation when necessary. In active mode, the controller allows proactive boundary adjustment by relaxing or tightening the boundaries based on predefined tasks or predictable conditions, delivering greater flexibility in performance management. To accommodate arbitrary initial errors, the initial performance boundary is set to infinity, and a new transformation function is introduced to overcome the issue of initial control failure. Unmeasured states are estimated via an observer, while unknown nonlinearities are approximated using fuzzy logic systems (FLSs). Simulation results demonstrate the proposed method’s effectiveness and superiority.
This article proposes a singularity-free prescribed performance control (PPC) scheme for nonlinear systems with actuator faults involving unknown positive odd integer powers. Unlike conventional fault-tolerant control schemes, where the exponent of the faulty input is fixed to 1, this work generalizes it to an unknown positive odd integer, directly coupling faults with higher-order input power characteristics. Moreover, departing from existing PPC methods, the study employs a Vandermonde matrix to construct a prescribed trajectory, transforming the original tracking error constraints into constraints on the deviation between the tracking error and the prescribed trajectory. This approach prevents input divergence even when the tracking error approaches the constraint boundary, effectively resolving the singularity issue observed in existing barrier function-based PPC schemes. Simulation results validate the effectiveness and superiority of the proposed scheme.
This article proposes a variable-gain extended state observer (VGESO)-based robust model predictive control (RMPC) framework for quadrotor trajectory tracking under nonvanishing disturbances. First, the VGESO is designed to estimate matched disturbances, which leverages its time-varying gain characteristic to achieve rapid transient estimation and maintain robustness against uncertainties. Furthermore, a model predictive control (MPC)-based control architecture is developed to achieve optimal trajectory tracking with explicit system constraints. Unlike existing quadrotor control approaches, the proposed scheme accounts for both matched and unmatched disturbances, and rigorous theoretical analysis is provided to guarantee recursive feasibility and closed-loop stability. Finally, the effectiveness and practicality of the proposed method are demonstrated through high-fidelity Gazebo simulations and practical experiments.
The decision variable grouping approach has proven effective in solving large-scale multiobjective optimization problems (LSMOPs). However, expensive large-scale multiobjective optimization poses additional challenges because the number of allowable function evaluations (FEs) is highly limited, making it difficult to perform decision variable grouping without relying on additional real FEs. To overcome this limitation, a novel entropy-informed variable grouping method that quantifies the distribution characteristics of each decision variable based on previously evaluated solutions is proposed in this study. In particular, the entropy of each decision variable is calculated to rank the variables. Those with similar entropy values are grouped to ensure that the grouping process accurately reflects the inherent distributional characteristics of the decision variables. The entropy-informed grouping results are then leveraged to guide the search process by prioritizing convergence for decision variables with higher entropy while focusing on diversity for those with lower entropy. Experimental evaluations demonstrate that the proposed method outperforms six state-of-the-art surrogate-assisted evolutionary algorithms (SAEAs) in both computational efficiency and solution quality, offering a robust solution for expensive LSMOPs.
Chaotic maps have been widely used in a variety of industrial applications, and their dynamic properties determine the performance of chaos-based applications to a large extent. However, designing chaotic maps with desired dynamic properties is a challenging topic. In this article, we propose a novel $n$ -dimensional discrete chaotic model ( $n$ D-DCM). By utilizing specially designed nonlinear operators and coupling modes, the proposed model can generate many enhanced chaotic maps with arbitrary dimensions. Based on the Lyapunov exponent (LE), the enhancement of chaos in both one and high dimensions is theoretically proved for our model. To further show the effectiveness of $n$ D-DCM, three illustrative chaotic maps with different dimensions are generated, and the performance evaluations show that our maps have outstanding dynamic properties. The microcontroller-based hardware implementations also indicate that our chaotic maps have good consistency and feasibility in hardware devices. Moreover, to show the potential of $n$ D-DCM in practical applications, we design a scheme for a pseudorandom number generator (PRNG). The evaluation results demonstrate that this scheme matches well with the proposed model, and our generators can efficiently produce pseudorandom sequences with high randomness.
Achieving safe, efficient, and human-like autonomous driving (AD) in complex real-world environments remains a critical challenge. To address the limitations of conventional AD learning paradigms that struggle with imperfect human demonstrations and insufficient safety guarantees, this article proposes a novel physics-informed and human-guided reinforcement learning (PI-HGRL) for AD, combining the dual advantages of human-like behavior from demonstrations with the safety assurances of physical models. PI-HGRL employs a dynamic, risk-aware task delegation mechanism that adapts to complex driving contexts by prioritizing human guidance in low-risk scenarios and relying more on physical models in high-risk situations, enhancing data quality and ensuring reliable, efficient human–AI collaboration. Moreover, a progressive fading intervention mechanism (PFIM), inspired by scaffolding theory in human education, is first introduced to systematically reduce reliance on hybrid enhanced guidance (HEG) and facilitate the agent’s transition from imitation learning to surpassing learning. Extensive experiments demonstrate that PI-HGRL outperforms baseline methods in driving safety, data efficiency, and generalization capability. Ablation studies further validate the contributions of the physical model and PFIM to the overall strategy. These results suggest that PI-HGRL offers a promising direction for building trustworthy and adaptive human–AI collaborative driving systems.
Wideband disturbances and measurement noise degrade pneumatic precision regulation, while high-gain extended state observers (ESOs) face an inherent tradeoff between disturbance reconstruction and noise amplification. This article proposes a hybrid adaptive extended state observer network (HAESO-Net) with two complementary branches, one emphasizing low-frequency disturbance reconstruction near resonances and the other prioritizing high-frequency noise attenuation. Their outputs are continuously fused through a residual high-frequency-energy index and a convex scheduling law, avoiding the limitations of fixed-structure selection or hard switching. Atracking-differentiator-based interface further regularizes adaptive-parameter and fusion-weight updates, reducing noise-driven gain excitation and observer-controller interaction in real-time implementation. Based on the fused state and disturbance estimates, a disturbance-feedforward adaptive backstepping integral sliding-mode controller is developed, and closed-loop bounded regulation is established. Experiments demonstrate approximately −30-dB attenuation in the 1–4-Hz band and rms reductions of 84.48% and 83.28% under 5- and 10-Hz resonance excitations; during a 1–100-Hz swept-sine with time-varying disturbance frequency, the scheduler adapts online and maintains wideband rejection as spectral dominance shifts.
This article investigates the problem of asynchronous attack-compensated control for discrete-time nonlinear semi-Markov jump systems (S-MJSs) subject to false data injection (FDI) attacks. To mitigate the adverse effects of malicious data, a resilient controller is constructed by combining a false-signal observer with a compensation-based control law. Concurrently, considering limited network resources, a memory-based adaptive event-triggered scheme is proposed to enhance control performance while significantly reducing communication overhead. Recognizing the practical constraints in transition information identification, the semi-Markov kernel (SMK) and the high-level homogeneous Markov chain are assumed to be partially available. By employing a mode-rule-dependent Lyapunov function together with the linear matrix inequality method, sufficient conditions are derived to guarantee the $H_{\infty }$ performance of the control systems. Finally, a single-link robot arm model is employed to validate the efficacy of the proposed compensation control strategy.
Meta-transfer learning introduces domain adaptation techniques into meta-learning frameworks, aiming to adapt to scenarios where training/testing tasks have both task distribution differences and domain distribution differences. However, existing meta-transfer learning methods tend to follow a metric-based meta-learning paradigm. Facing challenges such as data imbalance, task heterogeneity, and knowledge sharing in cross-domain few-shot hyperspectral image (HSI) classification. To address the above challenges, we proposed a balanced meta-adaptation network (BMAN) by customizing a meta-transfer learning framework for the cross-domain few-shot HSI classification task. First, to cope with the data imbalance challenge, we broke the traditional optimization mechanism in which the source domain and target domain are independent of each other. An inner optimization mechanism based on meta-generalized empirical risk minimization was constructed to capture the target domain-oriented few-shot classification knowledge. Second, to address the challenge of task heterogeneity, a balanced meta-sampling mechanism was designed to align the two domain task distributions, and an outer layer optimization mechanism capable of capturing unbiased knowledge was constructed to correct the feature learning pReferences of the model. Finally, to cope with the challenge of knowledge sharing, based on the idea of “seeking common ground while reserving differences,” we allowed the source domain to retain domain-specific knowledge while learning sharable knowledge, thus promoting cross-domain knowledge transfer. Experiments have shown that BMAN has achieved state-of-the-art performance in cross-domain few-shot HSI classification tasks.
In this article, we focus on addressing the vibration suppression and consensus tracking control problems for multiagent systems (MASs) composed of multiple moving vehicle-mounted flexible manipulator systems. Each agent is described by a partial differential equation (PDE). Suitable integral barrier Lyapunov functions (IBLFs) are constructed to guarantee that the output constraints are not violated. In addition, to address actuator failures at the boundary, unknown actuator faults are compensated using adaptive parameter estimation techniques. Subsequently, a distributed control protocol is designed to suppress vibration while achieving consensus tracking of both angle and vehicle position. The extended LaSalle invariance principle is employed to verify the asymptotic stability of the closed-loop system. Finally, the numerical simulation results illustrate that the constructed control method is feasible.
Efficient and reliable train dispatching is essential for ensuring the safety and punctuality of high-speed railways (HSRs) during emergencies. Current train dispatching relies heavily on manual operations performed by dispatchers and is governed by stringent operational procedures and safety regulations that require strict adherence to established rules. These complex tasks are highly prone to human error, especially in high-stress environments, and existing intelligent methods addressing specific aspects still require significant human intervention. To address these challenges, this article leverages the advanced comprehension, strategic planning, and coordination capabilities of large language models (LLMs) to introduce LLM-RailATD, an autonomous train dispatching method designed explicitly for HSRs during emergencies. LLM-RailATD operates through a structured four-stage approach-describe, plan, execute, and reason (DPER)-to interpret train operation scenarios, plan and execute complex train dispatching tasks, and autonomously handle errors arising during the dispatching process. Computational experiments based on real railway data illustrate the effectiveness of LLM-RailATD in autonomous train dispatching, which achieves a success rate (SR) of 74% in emergency scenarios. In addition, ablation studies validate the contributions of the individual modules of LLM-RailATD and the prompt design, highlighting their importance in achieving reliable performance.
Robust state-of-charge (SOC) estimation for LiFePO4 batteries depends critically on the accurate modeling of open-circuit voltage (OCV) hysteresis. However, the existing models fail to adequately capture the coupled effects of asymmetric hysteresis and temperature variations, leading to suboptimal model fidelity and unreliable SOC estimation. To address these issues, we propose an augmented nonlinear battery model integrating a temperature-compensated asymmetric hysteresis operator with cubic envelope functions, explicitly capturing charge/discharge asymmetry in a second-order equivalent circuit model. Model parameters are efficiently optimized across temperatures through metaheuristic-enhanced hierarchical identification. Building on this model, we develop a Bayesian-optimized adaptive sigma-point Kalman filter (SPKF) with self-tuning covariances. This mechanism enables automatic adaptation to OCV slope variations in different operating regions, ensuring robust SOC estimation by more precisely aligning Kalman gains with the underlying hysteresis dynamics. The experimental results under real-world applications demonstrate its superiority: the proposed model reduces modeling errors by 66.55% and 32.26% compared to two benchmark models, while the estimator maintains SOC error below 2% across diverse operating scenarios.
This work presents an event-triggered (ET) output feedback Lyapunov-based distributed model predictive control (DMPC) approach for large-scale nonlinear systems. In practical applications, incomplete state information, process disturbances, and measurement noise may degrade control performance. To address these issues, a distributed extended Kalman filter (DEKF) estimator is designed to reconstruct the system states for output feedback controller design. The convergence of the designed DEKF estimator is theoretically established. Based on the estimated states, an output feedback Lyapunov-based DMPC algorithm is developed by explicitly considering the influences of the coupling subsystems to reduce the scale of the control problem. Furthermore, an event-triggering condition is derived to reduce unnecessary online optimization and communication, forming a DEKF-based ET-DMPC framework. The recursive feasibility of the proposed ET-DMPC and the stability of the closed-loop system are rigorously proved. Finally, the proposed DEKF-based ET-DMPC algorithm is applied to a nonlinear continuous stirred-tank reactor (CSTR) system. The simulation results demonstrate that the proposed method reduces the computational burden while maintaining satisfactory control performance.
Multiple autonomous aerial vehicles (multi-AAVs) have demonstrated great potential in cooperative tasks such as target pursuit and flocking control. However, achieving efficient and adaptive multi-AAV cooperation remains challenging, especially in obstacle-cluttered environments. In this article, the hierarchical multiagent reinforcement learning with skill orchestration (HiMARS) approach is proposed to learn multi-AAV cooperative policies in complex environments. The core idea is to structure the learning process into a hierarchical architecture that integrates high-level skill orchestration and low-level skill learning. In particular, the complex cooperation task is decomposed into simple but complementary subtasks, where each low-level skill is separately trained to address a specific subtask. Then, a high-level skill orchestration mechanism is introduced to simultaneously leverage these skills for complex cooperation. By dynamically combining multiple skills, this mechanism not only enhances learning efficiency but also improves the adaptability of AAVs in complex environments. Finally, extensive simulations and real-world experiments in multi-AAV cooperative pursuit and flocking control tasks validate the effectiveness of the proposed HiMARS approach, demonstrating its superiority over state-of-the-art multiagent reinforcement learning (MARL) baselines.
