Series elastic actuators (SEAs) are widely used in human-robot interaction systems due to their intrinsic compliance and force sensing capability. However, nonlinear hysteresis in elastic elements can significantly degrade torque estimation accuracy and affect closed-loop force control performance. This paper proposes a hybrid modeling and control framework for SEA-driven hand exoskeleton systems. First, an optimized Archimedean spiral spring is designed to achieve compact torque sensing while satisfying stiffness and strength requirements. Second, a hybrid linear-LSTM hysteresis modeling method is developed for realtime torque estimation, where a physics-based linear model provides a stable recursive torque input to the neural network to improve numerical stability and hysteresis compensation accuracy. Third, based on the improved torque perception capability, a torque-sensing-based impedance control strategy is implemented for compliant human-robot interaction. Experimental results, including impedance control tests and preliminary clinical validation, demonstrate that the proposed method improves torque estimation accuracy, tracking performance, and interaction stability in SEA-driven exoskeleton systems.
Ding et al. (Fri,) studied this question.