Motivation and gap: PID-family controllers remain a pragmatic choice for many robotic systems due to their simplicity and interpretability, but tuning stable, high-performing gains is time consuming and typically non-transferable across robot morphologies, payloads, and deployment conditions. Fuzzy gain scheduling can provide interpretable online adjustment, yet its per-joint scaling and consequent parameters are platform dependent and difficult to tune systematically.

Proposed approach: We propose a hierarchical framework for cross-platform tuning of a learnable fuzzy gain-scheduled PID (LF-PID). The controller uses shared fuzzy membership partitions to preserve common error semantics, while learning per-joint scaling and Takagi–Sugeno consequent parameters that schedule PID gains online. Combined with physics-constrained virtual robot synthesis, meta-learning provides cross-platform initialization from robot physical features, and a lightweight reinforcement learning (RL) stage performs deployment-specific refinement under dynamics mismatch. Starting from three base simulated platforms, we generate 232 physically valid training variants via bounded perturbations of mass ($\pm$10%), inertia ($\pm$15%), friction ($\pm$20%), and damping ($\pm$30%).

Results and insight: We evaluate cross-platform generalization on two distinct systems (a 9-DOF serial manipulator and a 12-DOF quadruped) under multiple disturbance scenarios. The RL adaptation stage improves tracking performance on top of the meta-initialized controller, with up to 80.4% error reduction in challenging high-load joints (12.36$^\circ$$\rightarrow$2.42$^\circ$) and 19.2% improvement under parameter uncertainty. We further identify an optimization ceiling effect: online refinement yields substantial gains when the meta-initialized baseline exhibits localized deficiencies, but provides limited improvement when baseline quality is already uniformly strong.

It has always been expected that robots can actively manipulate complex environments to fulfill human requirements. This process typically necessitates that the robot be equipped with the ability for embodied exploration and manipulation. To achieve this goal, in this paper, we propose to incorporate multi-source knowledge to enhance the ability of robotic embodied exploration and manipulation. Specifically, to eliminate the inherent biases in decision-making of large language models (LLMs), we introduce a multi-source knowledge fusion module to generate more reasonable exploration sequences. Notably, grasping detection plays a critical role in the process of robot manipulation. To achieve a better balance between the accuracy and efficiency of the grasping detection network, we design a two-branch feature fusion module with residual blocks to improve network performance. Conditioned on the aforementioned innovations, the robot is capable of actively exploring and manipulating in constrained environments to meet human requirements. Extensive experiments are conducted in both simulation and real-world environments. The results demonstrate the effectiveness and efficiency of our proposed framework.

In this research, a hierarchical dynamic and kinematic modeling framework is proposed for a wheeled-legged manipulator (WLM), explicitly incorporating wheel slip and skid effects through the Gibbs–Appell formulation. Unlike traditional Lagrangian methods that depend on constraint multipliers, the proposed approach unifies the platform and manipulator dynamics while substantially reducing computational complexity. This integration enables efficient handling of nonholonomic constraints without compromising physical fidelity and offers a clear separation between subsystems, allowing the effects of wheel-ground interactions to be analyzed independently from manipulator motions. The proposed formulation is validated through a combination of MATLAB/Webots simulations and laboratory experiments conducted under both dry and wet (soapy ceramic) surface conditions. Experimental results indicate end-effector position deviations of 8–10.5% primarily due to wheel slip and initial joint torque discrepancies of 6.5–20% that progressively diminish as steady-state motion is reached. Comparative evaluation against a conventional Lagrangian model highlights the computational advantage of the Gibbs–Appell formulation, demonstrating reduced assembly time and fewer symbolic differentiation operations. Furthermore, a sensitivity analysis on friction coefficients and slip ratios confirms the robustness of the model to variations in surface conditions. Beyond accurate dynamic prediction, the hierarchical structure enables modular real-time implementation, supporting controller design, trajectory planning, and fault detection. Overall, the results demonstrate that the Gibbs–Appell-based hierarchical modeling framework combines analytical rigor with computational efficiency, providing a robust foundation for control and optimization in advanced wheeled-legged robotic manipulators.

With the rapid development of artificial intelligence technology, robotics, as its core branch, has attracted extensive attention from researchers. This paper designs and develops a robotic arm learning system based on multi-source sensor information fusion, which investigates the autonomous learning capability of robotic arms by closely integrating deep reinforcement learning (DRL) as the core framework for skill acquisition. By incorporating imitation learning as a source of expert prior and leveraging DRL’s intrinsic ability for policy optimization through environmental exploration, the proposed system achieves both rapid learning and robust generalization. Specifically, we introduce the gradient penalty mechanism from Wasserstein generative adversarial networks (WGANs), a technique that improves the stability of adversarial training by penalizing gradients that deviate from a specified norm. This mechanism is incorporated into the soft actor-critic (SAC) algorithm, a widely used off-policy DRL method known for its sample efficiency and robust performance in continuous control tasks. The resulting SAC-GP (SAC-gradient penalty) algorithm benefits from both SAC’s stable policy learning and WGAN’s improved training regularization, leading to superior convergence speed and system stability. Furthermore, this paper proposes a hybrid learning framework by combining generative adversarial imitation learning (GAIL) with SAC-GP, enabling the agent to benefit from both demonstration-based policy initialization and continuous self-improvement via reinforcement learning. Finally, a door-opening experiment is designed to verify the learning and execution capabilities of the system in both virtual and real environments. Experimental results demonstrate that the proposed learning system possesses excellent learning and motion execution abilities in practical applications. This achievement not only provides new insights for research in robot learning but also lays a solid foundation for the future development of robotic technology.

Upper limb rehabilitation exoskeletons face fundamental challenges achieving coordinated multi-joint assistance within portable configurations. Current cable-driven systems demonstrate effective single-joint support but lack coordinated shoulder-elbow capabilities due to anchor-point sensing constraints that limit workspace during simultaneous movements. To address these limitations, this paper presents a novel coordinated 3-Degree of Freedom (DOF) shoulder-elbow exosuit (3.3 kg) employing motor-proximal sensing architecture. Strategic load cell repositioning from anchor points to actuation unit locations eliminates spatial constraints, while geometric compensation algorithms maintain measurement accuracy, enabling coordinated assistance (shoulder flexion/extension, abduction/adduction, elbow flexion/extension) with preserved 0–$90^\circ$ kinematic workspace. Systematic development integrating product design specifications, multi-criteria decision-making, and biomechanical component dimensioning provided traceable design synthesis. Preliminary proof-of-concept tests with healthy participants (n = 5) provide an initial assessment of the system performance, demonstrating adequate sensing accuracy within the primary Activities of Daily Living (ADL) workspace, measurable muscle activation reduction, and preserved natural kinematics with minimal range-of-motion (ROM) constraint. An extensive experimental validation is out of the scope of this work. Results establish technical feasibility for coordinated cable-driven assistance, identifying requirements for future clinical translation.

Industrial inspection often requires systems to pass through confined entrances and traverse loosely constrained paths before reaching target locations. Borescopes and long-reach continuum manipulators are susceptible to insertion-induced buckling when deployed without continuous lateral support. This work presents a modular extra-long continuum manipulator designed for such hybrid industrial environments. The system combines a self-recoverable passive body with an independently actuated dual-balloon stabilisation unit, allowing it to traverse regions without continuous lateral confinement and subsequently establish a stabilised lift-up configuration at the target location. Its modular architecture supports transport and field maintenance, while the low-cost disposable design reduces decontamination requirements in harsh or contaminated environments. A solver-free, closed-form framework combines ground, tip, wall-contact, and wrapping resistances to estimate insertion-induced buckling margins for deployment planning. Prototypes (24 mm diameter, 8.2 m length) were validated experimentally and deployed in a nuclear decommissioning hot cell at Nuclear Restoration Services (NRS) Dounreay, UK, carrying a borescope camera and a 70 g radiation sensor to support inspection for subsequent decommissioning decisions.

This paper presents the design, analysis, dynamic dimensional synthesis, and prototype development of a novel 2-degree-of-freedom (2-DOF) parallel robot actuated by prismatic pairs. First, the proposed robot, designated as “TGU-SV2”, is introduced; it features prismatic actuation and spatial V-shaped passive limbs. This mechanism is designed to be modular, allowing integration with auxiliary modules to expand its DOFs. The forward and inverse kinematics, along with velocity and acceleration models, are derived to establish the Jacobian mapping relationships among the moving components. Based on these kinematic formulations, both simplified and exact rigid-body dynamic models are developed. The validity of these models is rigorously verified through co-simulations using SolidWorks and MATLAB/Simscape. Furthermore, three dynamic performance indices characterizing inertial forces, centrifugal/Coriolis forces, and gravitational terms are systematically formulated, subject to geometric and motion constraints. Monotonicity analysis and a genetic algorithm are then employed to determine the optimal dimensional parameters. Finally, a physical prototype is fabricated based on the optimized design, and preliminary performance tests are conducted. The test results show that the prototype can basically complete trajectory tracking, demonstrating its potential for applications in logistics sorting and industrial production, as well as laying a foundation for more refined prototype control in subsequent research.

To address the difficulty of simultaneously achieving motion bionics, active driving capability, and structural compactness in prosthetic knee joints, a structurally compact magnetorheological prosthetic knee joint (SCMRPK) with obstacle-crossing function is proposed in this paper. The SCMRPK combines a four-bar linkage mechanism and a linear magnetorheological actuator (LMRA). By integrating a micro linear motor into the piston rod of an MR damper, the LMRA combines the active and semi-active modes, enabling it to output driving force in the active mode and controllable damping force in the semi-active mode. The four-bar linkage mechanism is used to mimic the variation of the instantaneous center of rotation of the human knee. The structure and working principle of the SCMRPK are presented. A key dimensional design method for the LMRA magnetic circuit is proposed, and simulation and experimental tests are conducted to verify the effectiveness of the designed magnetic circuit. A control method for the SCMRPK is developed, and a prototype is fabricated. Under a locking current of 1.5 A, the LMRA prototype achieves stable displacement tracking in active mode. At a walking speed of 1.2 m/s, the semi-active mode of the SCMRPK prototype yields a root mean square error of 1.54° between the measured and ideal knee angles. In active mode, the SCMRPK-based lower-limb prosthesis successfully crosses an obstacle 0.3 m high. These results demonstrate that the proposed SCMRPK is compact and capable of achieving a relatively natural gait and obstacle crossing.

This paper proposes an indoor LiDAR-visual fusion framework based on dynamic elimination. The framework effectively suppresses transient dynamic interferences. It integrates multi-view geometric constraints with semantic priors, which is achieved through a prior model-guided dynamic feature elimination strategy. Moreover, an adaptive gamma correction (AGC) algorithm enhances the image quality under uneven illumination. Meanwhile, an improved quadtree algorithm homogenizes feature point distribution. These two measures achieve more robust ORB feature extraction while ensuring image quality. For map fusion, we focus on combining the processed visual map with the LiDAR map. The visual map, with dynamic feature points eliminated, is first aligned with the LiDAR map via Harris corner detection. We then integrate them using Bayesian inference-based grid information fusion to generate the final integrated map. Extensive evaluations were carried out on six dynamic sequence datasets, and practical dynamic corridor mapping experiments were also conducted. These evaluations and experiments demonstrate the superiority of the proposed system.

Greenhouse crop monitoring systems predominantly rely on sensor-based measurements of environmental conditions, such as temperature and humidity, while largely neglecting plant- and fruit-specific information. For delicate crops such as strawberries, this lack of plant-level perception limits the assessment of ripeness, spatial distribution, and localized growth conditions, which are essential for informed crop management. Enabling frequent, plant-specific visual and spatial monitoring therefore provides a necessary foundation for downstream tasks such as harvest planning and automated picking. To address this gap, this study presents a data collector robot (DCR) designed for plant-specific monitoring through a decoupled perception–action workflow. The proposed DCR integrates a red-green-blue-depth (RGB-D) camera, a robotic arm, and a rail-guided mobile platform to perform repeatable, multi-view image acquisition in greenhouse environments. Collected data are transmitted to a cloud server, where strawberry detection and localization are performed using a you-only-look-once (YOLO)-based model, and the processed results are stored for remote access and downstream use. This architecture enables accurate plant-level perception while avoiding the need for high-performance onboard computing. Experimental evaluations assess localization repeatability, robustness of cloud-based processing under variable network conditions, and continuous multi-plant operation. Results demonstrate a localization root-mean-square (RMS) of 0.39 mm, stable cloud-assisted inference with no dropped detections under simulated network impairments, and reliable performance across 100 complete data-collection cycles. A comparative analysis further shows that cloud-based processing offers a cost-effective alternative to local graphical processing unit (GPU)-based inference for scalable deployment. Overall, the proposed DCR provides a lightweight and modular solution for plant-level greenhouse monitoring and establishes a practical foundation for future autonomous strawberry harvesting systems.

Constant-force mechanisms (CFMs) are attractive for mechanical energy storage owing to their distinctive force–displacement characteristics, particularly under conditions with limited external load capacity and restricted space. However, conventional CFMs often suffer from short constant-force strokes and inefficient space utilization, which hinder their broader application. To address these limitations, this study exploits the buckling of compliant beams and increases the structural degrees of freedom by adopting a less constrained configuration, which extends the constant-force stroke and space utilization while reducing the required external load, thus improving energy storage efficiency for the same stored elastic energy. A novel catapult was developed through NSGA-II multi-objective optimization, achieving a high energy-to-cost ratio and an extended constant-force stroke. This work presents an effective design approach for motion mechanisms that demand high energy-storage efficiency and high-power output.

The shipborne 3UPS-3SS parallel mechanism is widely used in ship barge operation, but it has problems such as large volume and poor force transmission and acceleration performance. In order to solve the above problems, the kinematic/dynamic performance analysis and multi-objective optimization of scale parameters were carried out on the mechanism. The vector method is used to construct the position inverse solution model and constraint equations, and the kinematic inverse solution and velocity expression are derived, and the velocity Jacobian matrix is obtained. The fast polar coordinate search method was used to plot the reachable workspace by MATLAB software, and the volume expression of the workspace of the mechanism was obtained. The analytical equation of velocity and acceleration of each moving part was established by the differential method; the mechanism dynamics model was established by combining the virtual work principle, and the condition number was calculated and visualized based on the average mass matrix. Based on the distribution of the condition number of the velocity Jacobian matrix and average mass matrix, the force transferability and dynamic dexterity of the mechanism are evaluated. In the end, the workspace volume of the mechanism increased by 7.14%, the average minimum singular value increased by 196.43%, and the dynamic dexterity decreased by 55.69%, which provided a theoretical basis for the control and practical application of the mechanism.

To address the challenges posed by highly repetitive structures, unstructured terrain, and pronounced sensor disturbances in agricultural environments, this paper proposes a simultaneous localization and mapping framework that integrates explicit spatial feature modeling with time-varying noise estimation. First, a curved-voxel-based intensity–geometry joint probability model is constructed to transform conventional point features into spatial features capable of effectively capturing local structural patterns. These features are then aligned by minimizing the joint probability distance using the normal distributions transform, thereby enhancing feature discriminability and registration stability in highly repetitive scenes. Second, a maximum a posteriori-based recursive noise estimator is designed. By employing a dual sub-filter architecture, the proposed estimator enables joint modeling and online optimization of the noise parameters associated with LiDAR odometry and the inertial measurement unit, thereby improving the system’s adaptability to terrain-induced perturbations and sensor uncertainties. Experimental results demonstrate that the proposed method achieves a mean relative translation error of 1.42% and a mean relative rotation error of 1.53°/100 m in agricultural scenarios. In addition, the average error in row spacing estimation derived from the reconstructed map is constrained within 0.071 m. Compared with baseline methods, the proposed system exhibits significant advantages in both pose estimation accuracy and agronomic structural perception capability.

Multi-robot coverage path planning (MCPP) is a typical problem in multi-robot applications. Its goal is to decompose and assign coverage tasks, enabling mobile robots to collaboratively perform complete coverage of the map within a short time (referred to as coverage time). The coverage time of MCPP is generally determined by the path lengths of the robots and the number of turns on the paths. Most existing studies focus on only one of the two metrics, and few address the simultaneous optimization of both aspects. In this letter, we propose the Divide Areas based on area Growing via Wavefront Propagation (DAWG) algorithm to solve this problem through a staged co-optimization approach. Specifically, DAWG combines area growing and wavefront propagation to divide the map into equally sized subareas, ensuring equal path lengths. Meanwhile, DAWG incorporates a greedy strategy that reduces computation time by 81% while maintaining turn counts comparable to TMSTC*, a state-of-the-art (SOTA) method that minimizes path turns. Experiments on dozens of large-scale map instances demonstrate that DAWG achieves superior performance in both computation time and path quality compared to existing methods.

The appeal of quadruped robots lies not only in their ability to mimic the diverse and agile locomotion of animals but also in their potential to integrate with human artistic expression to achieve complex multi-skill movements. Continuous multi-skill motion in quadruped robots requires the realization of diverse, continuous, and long-horizon behaviors, involving a broader state space and more complex motion generation and transitions. This presents significant challenges, including sparse rewards, incomplete data, long-horizon motion planning, and the design of fine-grained motion transitions. In this work, we categorize quadruped robot skills into three types: rhythmic motions, expressive motions, and high-dynamic motions and generate reference trajectories for each category using central pattern generators, animation design, and motion capture, respectively. We then design an asymmetric neural network architecture and employ an imitation–reinforcement learning algorithm to train policies for generating these three types of motions. By composing multiple motion skill trajectories, we avoid long-horizon motion planning; by leveraging reinforcement learning, we enable smooth and continuous skill transitions; and by introducing a two-stage reference state initialization curriculum, the robot is able to switch from arbitrary states to the target motion skill. Moreover, during training, the policy imitates only key characteristics of the reference motions rather than strictly tracking fixed trajectories, making it more robust. Finally, we achieve robust motion skill generation and seamless transitions on a quadruped robot equipped with a 6-DoF manipulator, validating the effectiveness and feasibility of the proposed multi-skill generation and transition method.

The increasing adoption of robotic manipulators in the industry has brought the issue of robotic device safety to the forefront, particularly in the context of collaborative robotics. Among the main challenges, the evaluation of impacts with operators and objects in the workspace represents a critical issue, as accurately quantifying the magnitude of collisions is a complex task. Although current regulations propose various models for impact assessment, the suggested approaches are often oversimplified and not easily applicable in complex scenarios. To overcome this limitation, this study proposes the development of an effective mass impact model that simulates the collision between two bodies as an interaction between two lumped masses. The model is developed for both two-dimensional and three-dimensional analyses. Furthermore, it is implemented in a generic form and using a lumped-parameter robot model. This formulation enables the creation of a highly versatile model capable of overcoming the lack of knowledge of inertial parameters of robots. The approach is experimentally validated by means of a SCARA robot that collides with a cart, demonstrating the model’s applicability and accuracy.

Solid-state LiDARs (SSLs), due to their limited horizontal field of view (FoV), suffer from degeneracy in indoor environments. In response, we propose placing reflectors in degraded regions and incorporating reflector feature residuals into the LiDAR Odometry and Mapping (LOAM) framework. This is combined with an Extended Kalman Filter (EKF) to fuse odometry predictions, reflector observations, and LOAM measurements, effectively mitigating localization drift in the submap. By employing CostMap-Multilateration-based loop closure detection and keyframe selection strategies focused on high-reflection regions of interest (ROIs), our approach achieves robust submap-to-submap registration through a coarse-to-fine process. This ensures globally consistent map construction. To validate the proposed method, qualitative and quantitative evaluations were conducted in real-world environments. The results show that in degenerate indoor environments, the performance of our approach is comparable to, and even exceeds, that of state-of-the-art SLAM methods.

Neurological and musculoskeletal diseases have seriously impacted the quality of people’s lives. Various types of exoskeleton robots have been introduced to address this challenge in recent years. In this study, a lightweight and wearable underactuated cable-driven soft robotic intention-controlled glove exoskeleton (SRIG-Exo) is developed for assistance and rehabilitation training of individuals with injuries and disorders. The device comprises a soft robotic glove, a detachable cable connector, an underactuated linear actuator, and an active intention-controlled algorithm technique. The SRIG glove is equipped with tactile and flex sensors, along with tendons routed to facilitate finger movements. The detachable cable connector connects the glove with the actuator unit. The design of the linear actuator incorporates the threaded rod and nut mechanism to exert tension on the tendons. The intent signal from the glove sensors is utilized to activate the motor for opening and closing the hand, enabling active-controlled assistance and rehabilitation. This integration enhances the functionality and rehabilitation capabilities of the device. The constraints include the fingers having one degree of freedom (DoF), the control system requiring partial finger movement, and the wearer’s strength capability for normal device usage. The SRIG-Exo can support a maximum cable displacement range of 10 cm with a maximum grasping force of 30 N. Socially, this SRIG-Exo may improve independence in activities of daily living (ADLs), such as eating and drinking, for individuals with weakened strength and hand disorders, offering significant social and economic benefits.

In grid-based robotic path planning, traditional A* algorithms are often constrained by severe path redundancy, low search efficiency, and poor adaptability to complex environments with varying obstacle densities. To overcome these limitations, this article proposes the Hierarchical Adaptive Neighborhood Search A* (HANSA*) algorithm, which integrates a dual adaptive search mechanism with local environmental complexity evaluation. This mechanism enables dynamic switching between a 16-neighborhood pattern for rapid global exploration and a directional 5-neighborhood pattern for localized refinement based on real-time planning progress and local obstacle density. Phase-specific heuristic functions are introduced to balance search efficiency, path quality, and trajectory smoothness across different planning stages. Furthermore, a two-stage path optimization strategy is employed to reduce path redundancy, remove unnecessary waypoints, and alleviate sharp corners. Experimental results in dense and complex environments demonstrate that the HANSA* algorithm outperforms conventional approaches, exhibiting improved adaptability and robustness under varying obstacle-density distributions while generating paths with lower redundancy and fewer sharp corners. The results also show that HANSA* can replan paths after newly detected obstacles are incorporated into the updated occupancy grid map. These findings confirm the effectiveness of HANSA* for grid-based global path planning of mobile robots in complex environments.