Changes in muscle state in vertebrates directly influence their movement properties, offering inspiration for research on bionic robotic fish. This paper presents an untethered robotic fish equipped with a novel soft tail, which mimics the ability of real fish to modulate tail stiffness in real time. Inspired by the musculoskeletal system of organisms, the soft tail presented in this work primarily utilizes electrorheological fluid (ERF) and features a bionic structure mainly composed of a printed circuit board (PCB) with microfibers attached. Under the influence of a strong electric field, both the stiffness and damping of the tail increase simultaneously, thereby altering the hydrodynamic characteristics of the tail. Experiments demonstrate that the maximum increase in bending stiffness reaches 12% at a 5 kV voltage input. In dynamic tests, both thrust and swimming speed exhibit a strong correlation with stiffness. Under certain amplitudes and frequencies of the tailbeat, the swimming performance is effectively improved. Furthermore, active stiffness modulation facilitates progressive convergence toward the optimal Strouhal number regime observed in real fish. This study offers a novel approach to construct bionic robotic fish and highlights the effectiveness of stiffness adjustment in enhancing swimming performance.

Accurate real-time prediction of lower limb biomechanics is critical for enhancing assistive robotic control systems. While machine learning approaches have emerged as computationally efficient alternatives to traditional musculoskeletal multibody dynamics simulations, existing methods faced persistent challenges including limited prediction accuracy, inefficient multi-modal feature integration, and latency constraints inherent to current-time prediction frameworks. Therefore, this study aimed to propose a wearable sensors-driven deep learning model (KsFormer) for continuous multi-step ahead prediction of sagittal-plane joint moments (hip, knee, and ankle) and three-dimensional ground reaction forces (GRFs) across complete gait cycles. The encoder–decoder model KsFormer was specifically designed for cross-modal feature extraction and integration, with its encoder adopting a three-stage hierarchical processing pipeline. The preprocessed inertial measurement unit (IMU) kinematics and surface electromyography (sEMG) data recorded by wearable sensors served as the inputs of KsFormer. The prediction results were then validated by comparing them to those from gold-standard musculoskeletal simulations and force plate measurements. The results demonstrated exceptional predictive performance with mean Pearson correlation coefficients exceeding 0.9 across six walking speeds and three running speeds patterns, achieving low error rates ($\overline{\text{RMSE}}$ = 0.092 N m/kg for joint moments; $\overline{\text{RMSE}}$ = 0.032 body weight for GRFs). Additionally, the proposed model enabled accurate and continuous biomechanical prediction 240–960 ms prior to motion initiation, significantly outperforming conventional current-time prediction approaches. This study provided a more practical method for real-time lower limb biomechanics feedback to the assistive robotic system in the real-world environment, enabling dynamic torque adjustment and pilot gait pattern recognition.

Unlocking the high-performance potential of Hydraulically-Driven Soft Robotic Arms (HDSRAs) requires computationally tractable dynamic models that are both physically faithful and rigorously validated, a combination that remains a critical challenge. This paper addresses this gap by presenting a systematic framework for the modeling, identification, and multi-faceted validation of such systems. Central to the framework is an enhanced coupled dynamic model incorporating often-neglected physical phenomena, including stiffness coupling, Rayleigh damping, and pressure-dependent hydraulics. The framework’s value is then established through a cohesive suite of four targeted experimental studies. An ablation study first quantitatively confirms the necessity of each model enhancement. A comparative analysis subsequently demonstrates the model’s superior accuracy against representative existing methods. A model-based feedforward control experiment then proves the model’s practical utility by significantly improving trajectory tracking performance. Finally, a generalization study on a more complex tri-chamber arm confirms the framework’s scalability. This work delivers not just a model, but a fully validated, high-fidelity “digital twin” that provides a solid foundation for designing high-performance controllers for a broad class of HDSRAs.

Flexible needle puncture path planning in surgical robots faces significant challenges, such as limited adaptability to multi-objective environments and poor real-time performance. These issues affect both the accuracy and efficiency of needle puncture, restricting its application in complex medical scenarios. This paper proposes a deep reinforcement learning-based method to improve flexible needle path planning. To address the limitations of traditional models in accurately capturing needle dynamics, a hierarchical tissue model based on the unicycle framework is designed, which integrates kinematic and mechanical models. This approach considers the varying forces from different tissues on the needle at various positions. A dynamic multi-objective environment and obstacle model are also constructed, along with a target prioritization scheme for multi-objective optimization. Additionally, a prioritized experience replay (PER) mechanism is introduced to improve data efficiency in the learning process. This method enhances the needle’s adaptability and robustness in dynamic environments. Simulation results demonstrate that the model improved real-time performance and precision in dynamic multi-objective environments, making intelligent decisions based on target priorities and accelerating the exploration of optimal strategies.

Modern myoelectric prosthetic hands continue to face reliability challenges due to the non-stationarities of surface electromyography (sEMG) signals, which are highly sensitive to limb positions, electrode shifts, and grasp forces. While data abundance is a common strategy to mitigate these issues, it significantly increases users’ training burden. Hand gesture recognition, which maps the spatiotemporal patterns of muscle activation to key hand gestures for daily activities, remains the standard control strategy for advanced prosthetic hands. However, while temporal information can be reliably extracted from the sEMG signals, spatial information is highly dependent on electrode placements, which vary significantly between subjects. Previous research in myoelectric hand gesture transfer learning has primarily focused on transferring either spatial information or combined spatiotemporal information, leaving the transfer of temporal information alone largely unexplored. We propose a temporal-spectral cross-subject transfer learning framework using multi-stream convolutional neural networks (CNNs), where each stream processes only a single sEMG channel. Evaluated on the Transradial Amputee sEMG Multi-Contraction Forces Dataset, our framework has achieved training accuracy of 92.73% for medium contraction force and generalization accuracy of 74.53%, outperforming several models for sEMG hand gesture recognition. It also significantly improves recognition accuracy compared to the self-training baseline with the same architecture (repeated measures t-test p $\approx$ 0.032). By excluding spatial knowledge transfer, our approach maintains high robustness even under extreme cases of channel mismatch between source and target subjects. Moreover, this study highlights the importance of CNN architecture design, and spatially agnostic feature extraction for advancing myoelectric control systems.

Accurate recognition of gait phases and reliable prediction of gait cycles are critical for adaptive exoskeleton control. Although these two tasks are inherently coupled in both temporal and spatial domains, they have often been investigated in isolation. Such separation constrains the robustness and responsiveness of assistive devices in real-world applications. To bridge this gap, we propose a multi-task learning framework that integrates a Stacked Denoising Autoencoder with a Long Short-Term Memory network (SDA–LSTM). Inertial measurement unit (IMU) signals are employed for gait analysis, enabling both discrete gait phase classification and continuous estimation of gait cycle percentage during forward walking. The framework was evaluated against several established models, including Support Vector Machines (SVM), XGBoost, and standalone LSTM networks, under cross-subject validation. For phase classification, SDA–LSTM achieved 97.3% accuracy, with only a small reduction of 1.31% compared with its training performance. Robustness was further demonstrated under severe noise conditions (signal-to-noise ratio, SNR = 5 dB), where the model maintained 95.68% accuracy, surpassing all baseline methods. For cycle prediction, SDA–LSTM also showed strong stability. At SNR = 5 dB, error metrics increased only slightly, with RMSE and MAE rising by 0.083 and 0.075, while R² and PCC decreased marginally by 0.0173 and 0.0089. These results highlight the effectiveness of SDA–LSTM in capturing the spatiotemporal synergy of gait. The framework demonstrates high accuracy, robustness, and generalization. Its performance underscores strong potential for deployment in exoskeleton systems, paving the way toward reliable and adaptive human–robot interaction in daily locomotion.

Respiratory motion, as a dominant biological factor, imposes two primary challenges in radiotherapy: firstly, ensuring accurate spatial targeting of the tumor volume, and secondly, constraining radiation beams to minimize collateral damage to adjacent healthy tissues. Therefore, it is necessary to build a simulation platform for respiration tracking study before performing live surgery. In our previous work, a thoraco-abdominal respiratory motion phantom was proposed for simulating motion of chest, abdomen and internal tumor, which is helpful for respiratory motion tracking experiment. This paper validated that the phantom can reproduce the correlated respiratory movement of skin surface and tumor through experiments and analysis, which proves that the phantom can be used as an experimental components for the respiration tracking study of robotic radiosurgery.

Human legs exhibit spring-like behavior during the walking stance phase, motivating the development of lightweight passive lower extremity exoskeletons with elastic energy storage. While such designs offer promise in improving gait economy and reducing muscular effort, they often compromise swing-phase kinematics. Here, we propose a quasi-passive knee exoskeleton equipped with Clutched Elastic Actuators (CEAs) to assist human walking. The CEA on the knee joint comprises a magnetorheological fluid (MRF) bearing unit that serves as a clutch mechanism and a torsional spring for energy recycling. When the spring is engaged, the CEA stores energy during knee flexion and releases it during knee extension. Then, the spring is disengaged to permit unrestricted limb motion and maintain natural gait kinematics during the swing phase. Benchtop tests confirm that the MRF bearing unit delivers sufficient locking torque and enables smooth, rapid engagement and disengagement of the spring. Furthermore, with the assistance of the knee exoskeleton, treadmill walking experiments demonstrate notable reductions in muscle activity. Our approach paves the way for developing lightweight, inexpensive, and quasi-passive exoskeletons that reduce muscular effort and make recreational walking more enjoyable.

This paper presents a unified teleoperation framework for heterogeneous master–slave robotic systems, integrating geometric mapping with learning-based temporal modeling to enable skillful and adaptive manipulation. To address structural asymmetry and dynamic task variability, the proposed framework introduces four key components. (1) A DHT_xLSTM-based temporal modeling framework is proposed to enable context-aware skill prediction from multi-source sequential data, supporting autoregressive reproduction and long-horizon manipulation. (2) A unified master–slave mapping scheme is established by combining task-space pose alignment and joint-space transformation, enabling skill transfer across structurally asymmetric systems. (3) A unit dual quaternion-based joint-space mapping algorithm is introduced to ensure consistent directional transfer between mismatched human and robot joints, preserving motion semantics. (4) A dynamic hybrid control strategy is designed to switch between geometric mapping and learning-based prediction based on task phase, enabling seamless transition from gross to fine manipulation. Experimental results demonstrate that the proposed framework achieves high spatial fidelity, robust temporal generalization, and autonomous transition capabilities, laying a solid foundation for intelligent human–robot collaboration in complex manipulation tasks.

This study proposes a step adaptation framework for running through spring-mass trajectories and deadbeat control gain libraries. It includes four main parts: (1) Automatic spring-mass trajectory library generation; (2) Deadbeat control gain library generation through an actively controlled template model that resembles the whole-body dynamics well; (3) Trajectory selection policy development for step adaptation; (4) Mapping spring-mass trajectories to a humanoid model through a whole-body control (WBC) framework also accounting for closed-kinematic chain systems, self collisions, and reactive limb swinging. We show the inclusiveness and the robustness of the proposed framework through various challenging and agile behaviors such as running through randomly generated stepping stones, jumping over random obstacles, performing slalom motions, changing the running direction suddenly with a random leg, and rejecting significant disturbances and uncertainties through the MuJoCo physics simulator. We also perform additional simulations under a comprehensive set of uncertainties and noise to better justify the proposed method’s robustness against real-world challenges, such as signal noises, imprecision, modeling errors, and delays. All the aforementioned behaviors are performed with a single library and the same set of WBC control parameters without additional tuning. The spring-mass and the deadbeat control gain library are automatically computed in 4.5 s in total for 315 different trajectories.

Medical image segmentation takes an important position in various clinical applications. 2.5D-based segmentation models bridge the computational efficiency of 2D-based models with the spatial perception capabilities of 3D-based models. However, existing 2.5D-based models primarily adopt a single encoder to extract features of target and neighborhood slices, failing to effectively fuse inter-slice information, resulting in suboptimal segmentation performance. In this study, a novel momentum encoder-based inter-slice fusion transformer (MOSformer) is proposed to overcome this issue by leveraging inter-slice information from multi-scale feature maps extracted by different encoders. Specifically, dual encoders are employed to enhance feature distinguishability among different slices. One of the encoders is moving-averaged to maintain consistent slice representations. Moreover, an inter-slice fusion transformer (IF-Trans) module is developed to fuse inter-slice multi-scale features. MOSformer is evaluated on three benchmark datasets (Synapse, ACDC, and AMOS), achieving a new state-of-the-art with 85.63%, 92.19%, and 85.43% DSC, respectively. These results demonstrate MOSformer’s competitiveness in medical image segmentation.

Autonomous exploration in complex 3D environments is a key issue in robotics, where current approaches often demonstrate limited efficiency and excessive path backtracking. To mitigate backtracking and repeated exploration in complex multi-channel environments, we propose MCVP (Multi-Channel Viewpoint Planner), an autonomous exploration strategy consisting of three key components: viewpoints generation, viewpoints optimization, and dual-resolution exploration path generation. Firstly, MCVP employs a mixed-cost heuristic function to generate high-quality viewpoints by integrating key factors, such as distance, yaw angle, and positional constraints. Subsequently, a viewpoints optimization process is applied to eliminate redundancies and enhance computational efficiency. To establish an efficient mapping between viewpoints and channels, a bidirectional hash table structure indexed by distance-based criteria is utilized, enabling rapid correspondence retrieval. Finally, the system generates a dual-resolution exploration path, enabling efficient and adaptive navigation for mobile robots in complex environments. We evaluate the proposed method against state-of-the-art approaches in multiple challenging simulation scenarios. The quantitative and qualitative results demonstrate that our method can successfully achieve complete exploration across diverse environments with high efficiency , while exhibiting significant advantages in terms of exploration time and movement distance. To further validate the proposed approach, we conduct real-world experiments in both an underground parking and a complex university campus. The experimental results also further confirm the robustness and practical feasibility of our method in realistic unknown environments.

As industry transitions from the Industry 4.0 paradigm to the human-centered vision of Industry 5.0, robots are evolving from mere automation tools into intelligent partners capable of close collaboration with humans. Within this context, dual-arm human–robot collaboration (DA-HRC) has emerged as a key technology due to its human-like dexterity and ability to perform complex tasks. However, integrating dual-arm systems into collaborative environments is far more complex than merely doubling the number of manipulators; it also introduces a leap in system complexity, including challenges such as kinematic decoupling, dynamic coupling, and safe coordination. Despite growing research efforts, a comprehensive review of this inherently interdisciplinary field is still absent. This paper systematically surveys the literature on DA-HRC, with a particular focus on three core aspects: control, learning, and perception. It further discusses practical applications in industrial manufacturing and domestic service scenarios. The review highlights the ongoing paradigm shift from model-driven to data-driven approaches and identifies the integration of verifiable safety guarantees from traditional control theory with the adaptability of modern learning methods as a central technical challenge. Achieving safer, more efficient, and more natural interactions in dual-arm robots requires continuous iteration and deeper integration of these approaches.

Redundant manipulators possess additional degrees of freedom that enable superior dexterity, adaptability, and fault tolerance in complex environments. However, this redundancy also introduces challenges in inverse kinematics (IK) and redundancy resolution, as multiple feasible joint configurations may exist for a given end-effector task. This review systematically summarizes the state of the art in IK methods and optimization objectives for redundant manipulators. It first classifies IK approaches into analytical, numerical, optimization-based, and artificial intelligence-driven categories, highlighting their mathematical foundations, computational efficiency, and real-time feasibility. Next, various optimization objectives are analyzed from three perspectives: manipulability and dexterity indices, joint-level criteria such as torque or energy minimization, and task-level performance metrics including accuracy, smoothness, and collision avoidance. The integration of IK and redundancy resolution within hierarchical control, task-priority, and multi-objective frameworks is discussed, along with representative applications in industrial, medical, service, and space robotics. Finally, emerging research directions are identified, including hybrid learning-optimization paradigms, system-level fusion, collaborative redundancy resolution in multi-agent systems, and digital twin-enabled evaluation for trustworthy deployment. This survey provides both theoretical and practical insights for developing adaptive, explainable, and deployable redundancy-resolution systems for next-generation intelligent robots.

Lower limb exoskeleton technology has emerged as pivotal equipment in three key domains: medical rehabilitation, industrial assistance, and human performance augmentation. The systematization and standardization of evaluations regarding machine performance and human–robot interaction (HRI) are core challenges in transitioning this technology from laboratory research to mature applications. However, evaluation metrics emerging over the past decades have remained fragmented and lack systematic organization. Therefore, this review summarizes the mechanical structures and actuation modalities of common lower limb exoskeletons across these three domains. Starting with an elucidation of the specific evaluation priorities in each field, we review existing metrics – with a particular focus on metabolic cost, data precision, and human–robot compatibility – and construct a multi-level evaluation framework covering three dimensions: the human, machine performance, and HRI. Finally, this paper identifies current potential research focal points and overlooked areas, aiming to provide directional guidance for the future optimization of exoskeleton performance.