Most prosthetic hands cannot automatically control their gripping motions according to target objects; users need to manually select gripping motion. To address this problem, this study integrated a vision system with the prosthetic hand to autonomous...
Most prosthetic hands cannot automatically control their gripping motions according to target objects; users need to manually select gripping motion. To address this problem, this study integrated a vision system with the prosthetic hand to autonomously detect an object to grip and conduct a suitable grip motion. The users vision was replaced with a laser pointer and mini-cam. The user needs to indicate a target object with the laser pointer. Then, the object detection model (Scaled-YOLOv4) discerns the object class and determines the gripping motion corresponding to the class. To improve the object detection accuracy, various image processing methods are applied to the detection model. Moreover, a public robot hand design (Handi-hand) was significantly modified to realize precise and robust grip motions. Actuators, joints, and stiffness components in the robot fingers were replaced with different devices for high gripping performance, and the part design was changed accordingly. The unmovable wrist was also changed to an active wrist that is controlled by a servo motor. The gripping time was determined by the user’s electromyography signals. In experiments, this prosthetic hand successfully grasped four different objects.