Scholarlink

Patients who have had a stroke commonly develop upper limb motor dysfunction, particularly hand dysfunction, which significantly reduces their ability to perform Activities of Daily Living(ADL). As a neuromodulation technique, functional electrical stimulation (FES) offers a novel approach to enhancing hand-motor function. However, sophisticated hand motor control, customized stimulation program creation, and intelligent rehabilitation training are all lacking in current FES systems. This work proposes a hybrid reinforcement learning approach that uses pre-training and truncated proximal policy optimization (Truncated PPO) to build a FES based hand rehabilitation control system. A dataset representing the state-action correlation between hand joint motion and muscle stimulation is created by combining 24-channel flexible electrodes with high-precision Leap Motion sensors. The system’s ability to regulate hand movements with high precision and smoothness on a virtual hand model is demonstrated by experimental findings, which also confirm the efficacy of the proposed approach by greatly enhancing training convergence speed and state similarity.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Reinforcement Learning Control Method Based on Pre-training and Truncated PPO for Hand Rehabilitation

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Lu Liu,
  • Lu Sun,
  • Benyan Huo,
  • Yanhong Liu,
  • Zan Zhang

摘要

Patients who have had a stroke commonly develop upper limb motor dysfunction, particularly hand dysfunction, which significantly reduces their ability to perform Activities of Daily Living(ADL). As a neuromodulation technique, functional electrical stimulation (FES) offers a novel approach to enhancing hand-motor function. However, sophisticated hand motor control, customized stimulation program creation, and intelligent rehabilitation training are all lacking in current FES systems. This work proposes a hybrid reinforcement learning approach that uses pre-training and truncated proximal policy optimization (Truncated PPO) to build a FES based hand rehabilitation control system. A dataset representing the state-action correlation between hand joint motion and muscle stimulation is created by combining 24-channel flexible electrodes with high-precision Leap Motion sensors. The system’s ability to regulate hand movements with high precision and smoothness on a virtual hand model is demonstrated by experimental findings, which also confirm the efficacy of the proposed approach by greatly enhancing training convergence speed and state similarity.