Methods to Improve System Identification in a Human-Exoskeleton System

Robotic assistive devices have been proposed as a costeffective, repeatable alternative to conventional physical therapy. Recent successes in rehabilitative robotics for poststroke gait therapy [1] are due to a focus on the patient’s active engagement in the therapy. One quantitative measure of active engagement, muscle activity, is a highly non-linear signal that varies greatly between subjects. Therefore, devices which target that measure would benefit from a personalized, adaptive strategy, which does not rely on assumptions of how the patient population interacts with that device.