Human-in-the-Loop Modeling and Bilateral Skill Transfer Control of Soft Exoskeleton

Abstract

Soft exoskeletons (exosuits) are expected to provide a comfortable wearing experience and compliant assistance compared with traditional rigid exoskeleton robots. In this paper, an exosuit with twisted string actuators (TSAs) is developed to provide high-strength and variable-stiffness actuation for hemiplegic patients. By formulating the analytic model of the TSA and decoding the human impedance characteristic, the human-exosuit coupled dynamic model is constructed. An adaptive impedance controller is designed to transfer the skills of the patient's healthy limb (HL) to the bilateral impaired limb (IL) with a mirror training strategy, including the movement trajectory and stiffness profiles. A reinforcement learning (RL) algorithm is proposed to optimize the robotic assistance by adapting the impedance model parameters to the subject's performance. Experiments are conducted to demonstrate the effectiveness and superiority of the proposed method.