Prosthetic hands with wrist joints often rely solely on electromyographic (EMG) signals, which can cause wearer fatigue and reduce decoding accuracy, especially during prolonged use. To address this, researchers from Tongji University have developed a vision-servo-based method that replaces EMG for wrist joint control, enabling precise target alignment before finger manipulation via EMG.
Key takeaways
- The system integrates a compact camera (envelope volume <1 cm³) into the prosthetic hand's 'web' area, with a distributed control architecture for edge deployment.
- Two image-processing approaches are proposed: a classical OpenCV-based method (keyframe detection, GrabCut segmentation, Canny edge detection) and a lightweight MobileNet-SSD network, pruned and quantized for STM32MP157 deployment.
- A Jacobian mapping from pixel coordinates to joint angles enables servo control of wrist flexion and rotation, with proximity sensor signals triggering mode switching between visual and EMG control.
- In experiments, the method achieved an average adjustment time of ≤1.81 s, a manipulation success rate over 70%, and recognition accuracy of at least 90%, demonstrating its potential to reduce user burden and improve long-term operation.
This work offers a practical pathway for integrating vision-based control into lightweight prosthetic hands, potentially enhancing autonomy and usability in daily tasks.
Source: 《机器人》期刊 (robot.sia.cn) · Published 2026-03-25 · “受限环境下基于视觉伺服的假肢手目标对准方法”