Model-based assistance-as-needed for robotic movement therapy after stroke

This paper extends an adaptive control approach for robotic movement therapy that learns deficiencies in a patient's neuromuscular output and assists accordingly. In this method, adaptation is based on trajectory tracking error and a model of unimpaired motor control forces. The controller presented here adaptively learns and fills the gaps in the patient's ability to generate inertial forces, instead of just static forces, as has been proposed before. To test this method, a two dimensional model of an impaired human arm was used to simulate reaching movements in the horizontal plane. The results from simulation demonstrate that the inertia-based controller assists more effectively without need for increasing the controller's impedance, which suggests that modeling inertial forces during robot movement therapy could improve the ability of robots to deliver assistance-as-needed.