Heuristic-based online adaptation of ankle exoskeleton assistance using plantarflexor electromyography

INTRODUCTION People often change their coordination strategies as they learn to walk with ankle exoskeletons [1], yet most current exoskeleton control approaches do not appropriately account for these changes. Timebased assistance techniques, in which the exoskeleton is actuated at a specific point in the gait cycle [2, 3], keep device behavior static regardless of human adaptation. Proportional myoelectric control, in which torque provided from the exoskeleton is directly proportional to the user’s muscle activity [4], requires a certain level of muscle activity be maintained for the exoskeleton to supply torque, preventing muscle activity from being fully supplanted. Control techniques that adjust exoskeleton behavior online in response to measured changes in human coordination patterns would allow the human and device to co-adapt and potentially result in improved human-robot interaction. The goal of this project is to use soleus muscle activity, measured online, to update the desired ankle exoskeleton torque trajectory every step. Ideally, the exoskeleton will learn the changing pattern of soleus muscle activity over time and supplant the role the soleus muscle plays in plantarflexion. We expect to be able to tie our immediate goal, to drive soleus muscle activity down towards zero, to a higher level outcome, such as reduced whole-body metabolic rate.