EMG-Informed Musculoskeletal Modeling for Joint Stiffness Estimation Toward Rehabilitation Robot

Joint stiffness estimation of the human arm is a critical challenge for objective motor function assessment and human–robot interaction (HRI) of a rehabilitation robot. Perturbation-based methods are limited by their intermittent nature, whereas data-driven approaches often suffer from poor generalization and a lack of transparency. To bridge this gap, an electromyogram-informed musculoskeletal modeling (EMM) method is proposed to enable physiologically plausible estimation of joint stiffness. An analytical biceps–triceps antagonistic musculoskeletal model is developed, from which a muscle Jacobian matrix is derived to map contraction-aware muscle stiffness to joint space. Crucially, to ensure accurate personalization, a sensitivity-guided parameter tuning strategy is introduced, calibrating the model using only static isometric torque data without requiring extensive stiffness perturbation measurements. The method is validated through HRI experiments with nine participants and benchmarked against two subject-specific data-driven models. Results demonstrate that the EMM achieves high estimation accuracy (root-mean-square error (RMSE): 1.78 ± 0.41 Nm/rad, R²: 0.9763 ± 0.0089) and superior stability (lower error variance) compared to the benchmarks. Furthermore, the method successfully decouples the contributions of passive geometry and active neural drive to stiffness modulation, offering a robust, analytically efficient tool for rehabilitation robotics.