In context
In 2024, stroke rehabilitation robotics was advancing toward patient-active training paradigms, where human participation is critical for neural plasticity. However, controller design often neglected subjective intent, and transferring simulation-trained policies to physical robots posed safety and domain-shift challenges. This work addressed those gaps.
What was reported
Researchers at Hebei University of Technology proposed a reinforcement learning-based feedback control strategy for active upper-limb rehabilitation in stroke patients. They built an equivalent 6-DOF upper-limb model based on human anatomy and developed a random trajectory planning method using polynomial interpolation with Gaussian-distributed amplitudes and periods to mimic voluntary movements in joint space.
A simulation environment for the human-robot coupled system was constructed, mapping full system states to observable states via kinematic analysis. The control policy was trained with a reward function designed for active rehabilitation tasks. To transfer the policy to a real robot, a feedback control strategy aligned feature distributions between simulation and reality, preventing negative transfer.
Simulation results showed the policy could accomplish active rehabilitation tasks. Experiments on a physical upper-limb rehabilitation robot validated the learning and transfer method, demonstrating its effectiveness for active rehabilitation of movement disorder patients.
Why it mattered
This work provided a new technical solution for active rehabilitation by enabling safe policy transfer from simulation to real robots, addressing domain shift and safety concerns. It highlighted the potential of reinforcement learning in human-robot interaction systems, paving the way for more adaptive and patient-specific rehabilitation robotics.
"The control strategy trained in simulation is transferred to the real robot, and the experimental results verify the effectiveness of the learning and transfer method of rehabilitation strategy."
Source: 《机器人》期刊 (robot.sia.cn) · Published 2024-09-15 · “上肢康复机器人主动康复策略的学习及迁移方法”