Neuromechanics-based Deep Reinforcement Learning of Neurostimulation Control in FES cycling.

摘要: Functional Electrical Stimulation (FES) can restore motion to a paralysed person's muscles. Yet, control stimulating many muscles the practical function of entire limbs is an unsolved problem. Current neurostimulation engineering still relies on 20th Century approaches and correspondingly shows only modest results that require daily tinkering operate at all. Here, we present our state art Deep Reinforcement Learning (RL) developed for real time adaptive legs FES cycling. Core approach integration personalised neuromechanical component into reinforcement learning framework allows us train model efficiently without demanding extended training sessions with patient working out box. Our includes merges musculoskeletal models muscle or tendon multistate fatigue, render responsive paraplegic's cyclist instantaneous capacity. RL outperforms PID Fuzzy Logic controllers in accuracy performance. Crucially, system learned stimulate cyclist's from ramping up speed start maintaining high cadence steady racing as fatigue. A part has been successfully deployed Cybathlon 2020 bionic Olympics discipline paraplegic winning Silver medal among 9 competing teams.