Soft pneumatic robots have attracted widespread attention for their complete softness and strategic applications. They are controlled by rigid electronic pneumatic valves, significantly restricting their mobility and compliance. Most of the existing soft valves are driven by positive pressure, which may lead to leakage or even structural damage. Herein, a negative pressure-driven soft pneumatic valve (SPV) with binary input and output operations is proposed. It is convenient for SPV to expand the number of elastomer tubes passing through the valve chamber to further enhance the operation function. Additionally, eight logic gates are constructed by using SPVs, and their response of the input and output are measured. The logic gates can perform digital logic operations well. Then, higher level pneumatic digital processors are established. The higher level pneumatic digital processors can achieve more complex logic outputs, paving a new way for onboard control strategies of soft robots. The soft fluidic demultiplexer can control 23 actuators using three computer-controlled air pumps. Using the soft ring oscillator, the soft robot arm can be controlled by a constant negative pressure to rotate in a clockwise direction. The SPV can eliminate the need for additional electronic components, which takes a step towards electronic-free soft pneumatic robots.
This article presents the design, application, and experimental validation of an embedded actuation and control approach for vibration suppression in long-overhang turning tools. The proposed miniature device comprises an annular reaction mass with inner hybrid magnetic actuator, designed for integration at the free end of a boring bar without compromising tool functionality. The design incorporates an internal channel for coolant flow, providing thermal management of both the actuator and cutting tool. For ease of application, a data-based control synthesis method is developed for robust stabilization of tool vibration during cutting, effectively suppressing chatter caused by flexibility of the long tool. The control approach is based on a positive-real adaptive dynamic programming technique with off-policy iterations that use input–output data from the active device, supplemented by limited structural information. The methodology is validated through numerical simulations, hardware-in-the-loop experiments, and real turning tests. The results demonstrate a 95% reduction in resonance of the dominant tool mode, measurable improvements in surface roughness, and a marked extension in the cutting stability envelope—assessed via stability lobe diagrams.
Magnetically controlled microrobots hold promise for minimally invasive interventions in endovascular therapy; however, their navigation is hindered by strong pulsatile blood flow disturbances caused by the cardiac cycle. This article proposes a switching control framework for magnetically controlled microrobots to track vascular trajectories under periodic hemodynamic disturbances robustly. The controller switches between two modes: an active tracking mode during diastole (weak disturbance) and an antidisturbance mode during systole (intense disturbance), where an external magnetic field presses the microrobot against the vessel wall to suppress drift. For this hybrid system, a periodic Lyapunov analysis establishes stability conditions for closed-loop boundedness, derives an analytical upper bound on the steady-state tracking error, and predicts the cycle-level transient envelope toward an ultimate plateau. Experiments conducted on a pulsatile-flow test platform demonstrate that, compared to a nonswitching controller, the proposed strategy maintains stable navigation and improves tracking accuracy by an order of magnitude, validating theoretical predictions and highlighting its potential for safe endovascular operations.
Preserving mobility in the elderly and adults with chronic conditions is vital, as sedentarism leads to physical decline, disease, psychological issues, and increased fall risk. Traditional aids, such as walkers and canes, are common but they lack adaptability, comprehensive fall protection, and ease of use. This has driven interest in intelligent assistive devices. However, most existing robotic systems focus on therapist-led rehabilitation in clinical settings, limiting their use in everyday environments. We present intelligent walking assistive omni-directional exo-robot (I-WANDER), a novel robot designed to support, stabilize, and guide users during walking, standing, sitting, and other movements. I-WANDER combines innovative mechanical design with advanced control strategies to improve walking stability, enable shared control in therapy, and allow autonomous navigation for user following and target reaching. This study introduces the robot’s design and evaluates its performance via gait analysis, energy consumption metrics, and NASA-TLX assessments. Although experiments involved healthy participants, we simulated pathological gait to assess real-world applicability. Results show improved walking stability, reduced energy expenditure, and enhanced rehabilitation support. Our findings suggest that I-WANDER has huge potential to enhance user independence and safety through improved mobility and rehabilitation support.
Autonomous navigation in uneven and dynamic terrains requires reliable estimation of traversable free space and the ability to anticipate how moving obstacles will influence future motion. Traditional Agoraphilic* navigation provides free-space-driven steering but depends on cell-based geometric analysis, which is computationally intensive and sensitive to discretization. This article introduces an enhanced Agoraphilic* framework that integrates deep learning-based free-space estimation, automated data labeling, and predictive dynamic-obstacle reasoning to improve performance in real-world environments. A repurposed PointNet architecture is employed as a regressor that directly analyzes sectorwise point clouds to infer stability-aware free space without grid-based preprocessing. A fully automated labeling procedure is developed to generate sector-level free-space labels from terrain geometry and robot constraints, eliminating manual annotation and enabling scalable dataset creation. To support navigation in dynamic environments, a complementary prediction module estimates future free-space constriction by decomposing obstacle motion and evaluating a time-to-free-space-change. These static and dynamic free-space cues are fused within the steering formulation to produce efficient, geometry-aware, and responsive navigation commands. Real-world experiments in uneven terrains, including scenarios with moving obstacles, demonstrate that the proposed approach significantly reduces computation time, improves robustness to terrain irregularities, and enables reliable navigation in both static and dynamic settings.
Atomic force microscopy (AFM) is widely used in industrial nanometrology and scientific research. However, its high sensitivity to environmental and structural vibrations often introduces stripe-like artifacts, which compromise measurement accuracy. Conventional solutions—including passive isolation, image-domain denoising, and methods that attach auxiliary sensors to the sample or stage—are limited by mechanical path mismatch, extra vibration channels from the sensors themselves, and the risk of filtering out genuine surface features. To address these issues, this article presents an integrated Hinged dual-probe AFM. The design integrates an imaging probe and a squeeze-film damping-based vibration-sensing probe on a single mechanical structure, combined with a signal-to-noise ratio-guided differential denoising algorithm. Experiments conducted under both single-frequency and mixed-frequency vibrational disturbances show that the proposed system effectively suppresses periodic artifacts and improves measurement fidelity.
The dual-redundant electro-hydraulic brake (DREHB) system is of pivotal significance in ensuring vehicle braking reliability. However, a key challenge in leveraging the DREHB system lies in determining the redundancy switching time when faults occur. To address this issue, this article proposes a novel proactive switching control scheme for DREHB based on quantified fault-tolerant bound. First, a hybrid model prediction robust controller is designed for DREHB to achieve optimal performance under various system uncertainties. Second, the fault-tolerant bound of DREHB is numerically determined by constructing the maximal and minimal robust positively invariant sets. On this basis, a proactive redundancy switching mechanism is triggered to ensure timely switching prior to the occurrence of severe malfunctions. Simulation and experimental results validate the effectiveness and superiority of the proposed method.
Remaining useful life (RUL) prediction is critical for improving system reliability and reducing maintenance costs. However, significant discrepancies in feature distributions across different degradation stages hinder traditional methods from accurately identifying stage transitions. Moreover, during continual learning across degradation stages, models are prone to catastrophic forgetting, which undermines their prediction stability and generalization capability. To address these challenges, this article proposes a degradation-aware continual learning approach for cross-stage RUL prediction. First, a baseline guided gradient offset detection method was developed based on a constructed health indicator that characterizes degradation trends, which divides the training data into multiple degradation stages. Then, a stage aware residual neural network with enhanced residual convolutional network is trained on multistage data, which adaptively identifies the current degradation stage during online testing. Finally, to address the issues of cross-stage feature modeling and catastrophic forgetting, this article designs a cross-stage RUL prediction model based on the memory-aware adaptive transformation network, which effectively enhances the model’s ability to represent stage features and improves its continual learning performance. Experimental results on both battery and XJTU-SY datasets demonstrate the superiority of the proposed method in RUL prediction, highlighting its strong generalization ability and practical applicability.
Many industrial and field robotic applications, such as path-following for unmanned ground vehicles (UGVs), involve repetitive tasks that require high-precision tracking under physical actuator constraints. While classical iterative learning control (ILC) can achieve high accuracy when repeatedly following a given trajectory with a fixed completion time, its performance deteriorates when tracking a prescribed path whose execution time varies between trials. This article proposes a spatial iterative learning control (SILC) framework for path-following of a class of nonholonomic systems. A time projection mapping is introduced to transform trial-varying temporal dynamics into a unified spatial domain, enabling consistent learning and accurate path-following despite variable execution speeds. Rigorous convergence guarantees are established using singular perturbation theory and a composite energy function, which explicitly characterize how controller parameters can be tuned to achieve semiglobal practical asymptotic convergence while accounting for input saturation. Practical implementation issues, including discrete-time sampling and quantization effects, are systematically addressed. The proposed framework is validated through comprehensive simulations and physical experiments. Results obtained on tracked UGVs—a representative special case of nonholonomic systems—demonstrate superior tracking accuracy and robustness to unmodeled terrain disturbances compared with conventional feedback controllers, confirming both the theoretical rigor and practical viability of the proposed approach.
This article presents an adaptive haptic control framework for a dynamic manual wheelchair simulator designed for urban accessibility analysis. Unlike conventional simulators relying on fixed resistance models, the proposed approach continuously adapts to variations in human-wheelchair interaction dynamics. A seven-DOF reference model generates trajectories for straight-line, turning, slope, and cross-slope maneuvers, driving a motion platform and haptic ergometer through coordinated force feedback. The control architecture combines online friction estimation with a Takagi–Sugeno (T–S) fuzzy $H_\infty$ controller. Gradient-based adaptation laws estimate unknown contact friction parameters with Lyapunov-guaranteed convergence. Since T–S premise variables are unmeasurable, membership functions are reconstructed online from estimated parameters. Friction uncertainty, input saturation, and external disturbances are handled jointly; input-to-state stability is certified via linear matrix inequalities, with the small-gain condition numerically verified at a safety factor of 3.57. Experimental validation with seven participants across user weights (40–80 kg) shows tracking-error reductions of 58%–82% over MRAC and 65%–90% over model predictive control-linear parameter-varying at 100 Hz. Realism ratings reach 90%–95% for straight-line, slope, and braking scenarios, but drop to 45%–58% in cross-slope due to longitudinal-only friction modeling. Future work targets lateral friction modeling and platform synchronization.
In this article, we study the problem of remote state estimation for active suspension systems (ASSs) over relay channels. A novel consensus-based distributed estimation scheme is proposed, which achieves accurate state estimation in the presence of multiscale channel fading and disturbance constraints. To do this, distributed state and measurement models for ASSs are formulated. A relay-assisted architecture is then introduced to enhance communication reliability. Within this framework, a relay filter with tunable performance is designed to preprocess measurements at the relay node, followed by a distributed moving horizon estimator (DMHE) on the remote platform. The DMHE incorporates a new consensus strategy to fuse data from the relay and local nodes. Sufficient conditions are derived to guarantee the stochastic ultimate boundedness of both the relay error and the state estimation error. Numerical simulations and experiments on an electrically interconnected suspension platform demonstrate the effectiveness and superiority of the proposed scheme compared to existing methods.
Directional tactile feedback has been explored as a means of conveying robot-environment contact information to support spatial perception in remote manipulation systems. Delivering multidirectional feedback with a compact form factor remains a significant challenge. In this study, we propose a compact vibrotactile actuator that generates four-directional feedback via direct mechanical impacts. The actuator utilizes electromagnetic force to propel a neodymium magnet (slider) toward an impact layer. The slider returns to its initial position through the repulsive magnetic force exerted by a magnetic spring after impacts. The magnetic spring enables an untethered configuration that allows direct impact transmission in four distinct directions. The compliant impact layer, composed of soft corners and rigid impact blocks, effectively transmits the impact force only to the intended direction. In the user study, participants achieved 100% accuracy (95% CI: [98.5%, 100%], $p < 0.001$) in discriminating two-directional feedback during inferior pincer grasping, and 99.25% accuracy (95% CI: [97.8%, 99.8%], $p < 0.001$) in discriminating four-directional feedback during quadpod grasping. Furthermore, we demonstrate the teleoperation applications by providing directional feedback corresponding to collision locations detected at the end-effector of a robotic arm.
Direct collaboration between robots and humans requires control strategies that ensure safety without sacrificing efficiency. However, compliance with standards, such as ISO/TS 15066, often leads to overly conservative robot behaviors. This article proposes a safety-informed model predictive path integral (SI-MPPI) control framework that guarantees safety while preserving task performance. The architecture integrates a constrained MPPI planner and a trajectory scaling module. The first explicitly incorporates the safety constraints and adaptively computes the stochastically optimal sequence of safe waypoints in real time. Exploiting the receding horizon principle, it selects control actions that maximize long-term efficiency, deciding whether to reduce speed or deviate from the preplanned path. The second refines these waypoints via spline interpolation and velocity adjustment. The approach has been validated in both simulation and real collaborative tasks with a UR10e manipulator, showing fast adaptation to human motion and improved efficiency compared to a state-of-the-art method, while maintaining full compliance with safety requirements.
Hybrid rigid-flexible hand exoskeletons show great potential for hand function restoration and augmentation, but their practical use is hindered by joint misalignment, insufficient output force, and complex nonlinear dynamics. To address these challenges, this article presents a Bowden cable-driven rigid-flexible hand exoskeleton for high-performance grasping assistance. A compact self-aligning mechanism based on a cavity guide and a passive spring steel plate enables adaptive elongation to reduce joint misalignment and improve wearing comfort. Inspired by human tendons' stiffness regulation, a bionic exo-tendon is designed to enhance fingertip output force through passive tensioning as the finger flexion angle increases. Furthermore, an extended Kalman filter-based active model control (EKF-AMC) strategy is elaborated to estimate finger flexion states with the Bouc–Wen model, aggregate Bowden cable bending variations and hysteresis into a lumped disturbance, and compensate complex dynamics without specialized Bowden cable bending angle sensors. Experimental validation shows that the finger module weighs only 10.96 g and achieves a peak fingertip force of 14.1 N, and the self-aligning mechanism reduces peak misaligning force by over 60%. EKF-AMC consistently improves tracking performance at 0.125, 0.5, and 1 Hz, achieving up to 77.0% improvement compared with baseline controllers.
Industrial inkjet printing platforms require embedded intelligence that tightly integrates sensing, actuation, control, and embedded computing under strict constraints on computation, memory, energy, and deterministic timing. However, privacy and compliance restrictions prevent raw data sharing, limiting the effectiveness of cloud-based learning. Federated learning addresses privacy concerns through model-update sharing, but centralized orchestration suffers from server bottlenecks and single-point failures, creating communication congestion and scalability limitations in industrial deployments. Decentralized federated learning removes the central server, yet synchronous all-neighbor aggregation remains vulnerable to stragglers and model divergence under heterogeneous and non-independent and identically distributed (non-IID) conditions. To address these challenges, we propose Rising from Embodied Intelligence to Embedded Intelligence via Decentralized Federated Learning with Roaming AI Agents (REEDFL), a decentralized federated learning framework with roaming AI agents for industrial embedded intelligence. Instead of global synchronization, roaming agents carry model states and perform hop-by-hop training across the peer topology, activating only agent-hosting nodes in each round. When multiple agents meet at a node, a consistency-preserving multiagent fusion module weights their states according to compatibility with local data, providing stable initialization for subsequent training. An intelligent scheduling mechanism further selects next hops by jointly considering data utility, resource availability, and forgetting signals. Experiments on a real five-node industrial inkjet platform show that REEDFL achieves faster convergence and up to 5.06% higher accuracy than the strongest baseline.
Microscopic rotation plays a pivotal role in achieving precise alignment across diverse scientific and technological domains. To achieve submicroradian resolution, a flexure microscopic rotation converter that transforms translational input into refined rotation is proposed. By serially connecting a differential displacement reducer and a Scott–Russell inverter, the device inverts and significantly reduces translational input to generate high-precision rotation. A kinetostatic model was developed to optimize the converter’s parameters, specifically aiming to minimize conversion ratio variation. The conversion performances of the optimized design were validated by a finite element model. The prototype test shows that the average conversion ratio is 1 $\mu$rad/$\mu$m with a linearity of 0.44%. By adopting an internal model controller, the microscopic rotation converter achieves dynamic performance with tracking error less than 5.31% across the frequency range from 1 to 100 Hz.
Robust balance control is critical for quadruped robots to navigate unstructured environments and withstand external perturbations. This article presents a novel nonlinear model predictive control framework, termed simultaneous state and zero moment point (ZMP) tracking (SSZT) MPC, that unifies two distinct control paradigms in legged robotics: the state-tracking and reactive-stepping approach prevalent in quadruped control, and the explicit ZMP-based balance management traditionally used in humanoid robotics. This unification is realized with a cost function to be jointly optimized for centroidal state tracking and an input-derived ZMP computed directly from the optimized ground reaction forces. To further improve robustness, we introduce an online adaptation scheme that utilizes center of mass velocity errors to dynamically adjust the ZMP reference, creating a more achievable balance target in real-time. Validation on our full-size quadruped, LeoQuad, demonstrates that SSZT MPC consistently and significantly improves robustness over baseline approaches. Experimental results show superior performance across a wide range of challenging tasks requiring proactive balance management, such as external disturbance rejection, traversal of uneven terrain, stair climbing, and dynamic acceleration/deceleration.
