biomechanical analysis

This work addresses the challenge of quantifying human–robot coupled dynamics in physical interaction scenarios, where existing robotic designs often lack direct access to internal human biomechanical states such as muscle forces and joint loads. To overcome this limitation, the authors propose a simulation-based, scalable framework that integrates an embodied human musculoskeletal model as a physiologically plausible behavioral agent. By combining reinforcement learning controllers with pretrained human motor policies, the approach enables co-optimization of robotic morphology and control strategies. The framework facilitates quantitative assessment and joint optimization of internal biomechanical metrics within the human–robot system. Evaluated on human–exoskeleton interaction tasks, it significantly improves joint alignment and reduces contact forces, demonstrating its effectiveness for data-driven, biomechanically informed design of interactive robots.

To address safety risks associated with early-stage user testing in assistive and rehabilitation robotics, this study proposes an experimental framework integrating a hemiparesis-simulating suit with digital twin technology. Healthy participants wear a custom-designed suit to safely emulate impaired gait, while synchronized kinematic (Vicon), electromyographic (Delsys), and inertial data are acquired. Multimodal fusion and machine learning models characterize dynamic gait adaptation mechanisms and human–walker interaction patterns. The key innovation lies in the first closed-loop coupling of a physical simulation suit with a digital twin, enabling interpretable, dynamic representation of pathological gait. Experimental validation confirms that the suit significantly alters joint kinematics and muscle activation timing; a highly discriminative sensor feature set is identified, enabling robust detection of assistive device usage states and turning intent. This framework establishes a transferable, rapid-prototyping validation paradigm for rehabilitation robots.

Chronic pain often leads to diminished functional capacity, yet objective and convenient methods for quantifying patients’ motor function in real-world settings remain scarce. This study proposes a computer vision approach leveraging monocular smartphone videos and deep learning–driven 3D human pose estimation to remotely extract clinically relevant kinematic biomarkers in home environments without specialized equipment. Through systematic bias correction and individualized leave-one-subject-out calibration, the method demonstrates high agreement with gold-standard optical motion capture in laboratory validation (r > 0.85) and exhibits excellent test–retest reliability (r > 0.86) as well as significant sensitivity to group differences in patients with fibromyalgia and sciatica. This work represents the first demonstration of high-precision, scalable 3D motor function assessment in unsupervised home settings.

Existing musculoskeletal modeling and simulation approaches are hindered by reliance on expensive sensors, laboratory settings, high computational overhead, and poor tool integration. To address these limitations, this work introduces the first wearable, real-time musculoskeletal simulation framework that deeply integrates OpenSimRT with ROS, enabling online estimation of inverse kinematics, dynamics, and muscle activations. The system leverages low-cost wearable sensors—including IMUs, RFID tags, and pressure-sensing insoles—and employs ROS for low-latency, multi-source data communication and end-to-end real-time closed-loop analysis. Validation across daily activities—walking, squatting, and sit-to-stand/stand-to-sit transitions—demonstrates accurate estimation of ankle joint dynamics and key lower-limb muscle activations, confirming its efficacy in unconstrained, real-world environments. This framework overcomes environmental constraints, establishing a lightweight, deployable paradigm for real-time musculoskeletal analysis—advancing applications in rehabilitation assessment, prosthetic control, and exoskeleton design.

Existing whole-body musculoskeletal models suffer from limited muscle counts (<100) and joint degrees of freedom, hindering high-fidelity, real-time coordinated control of 600+ muscles. To address this, we propose MS-HUMAN-700—the first full-body musculoskeletal model featuring 700 anatomically grounded muscle units, 90 body segments, and 206 kinematic joints. We design a hierarchical low-dimensional representation framework that maps the high-dimensional muscle activation space into a learnable latent space. Integrating biomechanical modeling with hierarchical deep reinforcement learning, our method enables closed-loop, muscle-level motor control. In simulation, MS-HUMAN-700 accurately reproduces human gait patterns with state-of-the-art control fidelity. Both the model and algorithm are fully open-sourced, establishing a scalable neuro-muscular control foundation for embodied intelligence and human–machine interaction.

The humanoid robotics community lacks a standardized, quantitative benchmark for “human-level” joint actuation performance; conventional single-metric evaluations (e.g., peak torque) fail to capture the coupled constraints of torque, power, and endurance under task-relevant postures and velocities. Method: We propose the first quantifiable evaluation framework for human-level actuation, introducing the Human-Equivalent Envelope (HEE) and Human-Level Actuation Score (HLAS), grounded in the International Society of Biomechanics (ISB) standard for kinematic degrees of freedom. Comprehensive performance is measured across multiple joints under realistic operational conditions via dynamic torque testing, electrical power monitoring, and thermal sustainability experiments. Contribution/Results: Our framework uncovers fundamental trade-offs among gear ratio, bandwidth, and efficiency, enabling systematic, cross-platform comparison of actuation performance. It significantly enhances both the comparability and practical relevance of drive system assessment in humanoid robotics.

This study addresses the identification of neuromotor behavioral differences between healthy individuals and stroke patients during isometric upper-limb gaming tasks to inform the design of rehabilitation robot interfaces. Leveraging six-dimensional force, surface electromyography (sEMG), and task performance data from 13 healthy participants and 2 stroke survivors, the authors propose a hidden Markov model (HMM)-based dynamic sEMG classification method that significantly outperforms conventional muscle synergy decomposition approaches in distinguishing neuromotor patterns between the two groups. The findings reveal that task constraint axes and instruction comprehension substantially influence behavioral performance, with six-dimensional force data exhibiting significant intergroup differences (p = 0.05). These results validate the efficacy of HMM for neuromotor behavior recognition and provide empirical support for optimizing human–robot interaction in post-stroke rehabilitation systems.

Existing human–robot force interaction models for wearable robots rely predominantly on single-variable, single-degree-of-freedom force representations, failing to capture the nonlinear interfacial mechanical response of soft tissues under coupled normal and tangential loading. Method: We propose a bivariate force representation framework that jointly models normal and tangential forces, overcoming the limitations of conventional univariate fitting. Leveraging finite element simulations and soft-tissue mechanical experiments, we quantitatively evaluate the predictive accuracy of multiple constitutive models for force and torque across diverse loading conditions using normalized mean square error (NMSE). Results: The proposed method significantly reduces simulation error and improves fidelity in predicting pressure distribution and shear stress. It further identifies systematic biases inherent in univariate models under multi-degree-of-freedom interactions, establishing a new paradigm—supported by quantitative metrics—for high-fidelity interface modeling, material parameter identification, and optimization-driven design of wearable robotic interfaces.

This study addresses the limitations of existing lower-limb exoskeleton control methods, which often rely on extensive data or iterative parameter tuning and thus struggle to generalize to clinical populations. The authors propose a device-agnostic, personalized control framework that uniquely integrates physiologically plausible musculoskeletal simulation with reinforcement learning to automatically generate gait assistance strategies aligned with human biomechanics—without requiring task-specific parameterization. By unifying the modeling of both healthy and pathological gait patterns, the approach produces asymmetric assistive torques tailored to individual muscle weakness. Simulation results demonstrate that the generated hip and ankle torque profiles closely match those of state-of-the-art human-in-the-loop strategies, significantly reducing metabolic cost across multiple walking speeds and effectively improving energy efficiency and bilateral movement symmetry in simulated pathological gait.

This study evaluates the ecological validity of a real-time manual wheelchair simulator at both biomechanical and perceptual levels to ensure its fidelity in replicating real-world propulsion performance. Validity was systematically assessed by comparing kinetic data—collected via instrumented wheel systems—and subjective user experiences during straight-line propulsion, acceleration, turning, and ramp tasks performed both in the simulator and in real-world environments. The evaluation innovatively integrates temporal parameters, kinetic characteristics, waveform similarity of propulsion cycles, and multidimensional perceptual metrics, while accommodating users’ own wheelchair configurations. Results demonstrate high congruence between simulated and real-world performance in terms of timing and propulsion waveform morphology. Participants consistently endorsed the simulator’s safety and training utility, underscoring its potential as an assistive technology tool.