Compliant Control of Post-Stroke Rehabilitation Robots: Using Movement-Specific Models to Improve Controller Performance

Abstract

Post-stroke neurorehabilitation is an emerging application field of robotics, aiming to design new treatment systems and protocols based on the use of robotic technology and virtual reality to improve patient recovery after stroke. One goal in this field is to develop robotic therapy devices that are compliant but can still assist weakened patients in making desired movements. It is hypothesized that, in this way, the interaction with the robotic system can maintain patient engagement and effort, and promote and stimulate the motor learning process of the patient. One way that has been proposed to maintain compliance while assisting weak patients is to use an adaptive controller with a forgetting term, which allows the robotic system to learn a model of the forces needed to assist the patients during exercises while encouraging patient effort. A limitation of such an approach is that the adaptive gain must be large enough to rapidly change the model for different target movements, which decreases the compliance of the robot. We show here in simulation that by building independent models for different target movements, robot compliance can be increased while still accurately achieving the target movements.