Optimizing Compliant, Model-Based Robotic Assistance to Promote Neurorehabilitation

Abstract

Based on evidence from recent experiments in motor learning and neurorehabilitation, we hypothesize that three desirable features for a controller for robot-aided movement training following stroke are high mechanical compliance, the ability to assist patients in completing desired movements, and the ability to provide only the minimum assistance necessary. This paper presents a novel controller that successfully exhibits these characteristics. The controller uses a standard model-based, adaptive control approach in order to learn the patient's abilities and assist in completing movements while remaining compliant. Assistance-as-needed is achieved by adding a novel force reducing term to the adaptive control law, which decays the force output from the robot when errors in task execution are small. Several tests are presented using the upper extremity robotic therapy device named Pneu-WREX to evaluate the performance of the adaptive, "assist-as-needed" controller with people who have suffered a stroke. The results of these experiments illustrate the "slacking" behavior of human motor control: given the opportunity, the human patient will reduce his or her output, letting the robotic device do the work for it. The experiments also demonstrate how including the "assist-as-needed" modification in the controller increases participation from the motor system.