Reinforcement Learning for Hand Grasp with Surface Multi-field Neuroprostheses

Abstract

Hand grasp is a complex system that plays an important role in the activities of daily living. Upper-limb neuroprostheses aim at restor- ing lost reaching and grasping functions on people su ering from neural disorders. However, the dimensionality and complexity of the upper-limb makes the neuroprostheses modeling and control challenging. In this work we present preliminary results for checking the feasibility of using a re- inforcement learning (RL) approach for achieving grasp functions with a surface multi- eld neuroprosthesis for grasping. Grasps from 20 healthy subjects were recorded to build a reference for the RL system and then two di erent award strategies were tested on simulations based on neuro- fuzzy models of hemiplegic patients. These rst results suggest that RL might be a possible solution for obtaining grasp function by means of multi- eld neuroprostheses in the near future.