Design of joint torque feedback control for parallel ankle rehabilitation robot

To overcome the limitations of existing ankle rehabilitation algorithms that focus primarily on mechanical aspects while overlooking biological factors, this study proposes a joint torque feedback control (JTFC) algorithm based on a parallel ankle rehabilitation robot (3-SPS/RU*R configuration). The algorithm aims to prevent joint overload and secondary injuries caused by excessive driving forces. An inverse dynamics model integrating both mechanical and biological elements was developed using the virtual screw method. The JTFC system employs neural network prediction and a multi-objective genetic algorithm (MOGA) to optimize motion parameters, alongside particle swarm optimization (PSO) for weighted PID tuning, ensuring joint torque remains within safe thresholds. Experimental validation using a physical prototype demonstrated high tracking accuracy (NRMSE as low as 0.0002), robustness (R-values of 0.7495–0.7499), and energy efficiency (Ec: 17.6254–25.0527). The results confirm the algorithm’s precision, stability, and practical applicability, offering valuable theoretical and practical insights for biomechanically-integrated ankle rehabilitation.