Assist-as-needed control with a soft robotic glove based on human-object contact estimation

Assist-as-needed control with a soft robotic hand glove for active rehabilitation is studied in this work. There are two resources of the grasping force, the robotic glove and the subject. Compared with traditional passive rehabilitation where the grasping force is merely provided by a robotic hand rehabilitation device (such as hand exoskeleton, robotic glove), assist-as-needed control accounts for the user contribute to performing grasping tasks collaboratively. In this control method, the human muscle strength for grasping is estimated through the myoelectrical signals of the human forearm collected by the MYO armband. A neural network is used for the recognition of human-object contact estimation. The assist-as-needed control is finally implemented to assist humans in grasping tasks. Experiment results on a soft robotic glove show the effectiveness of the proposed assistive control method.