Trajectory Adaptation and Learning for Ankle Rehabilitation Using a 3-PRS Parallel Robot

This paper presents a methodology for learning and adaptation of a 3-PRS parallel robot skills for ankle rehabilitation. Passive exercises have been designed to train dorsi/plantar flexion, inversion/eversion ankle movements. During exercises, forces may be high because patient cannot follow the desired trajectory. While small errors in the desired trajectory can cause important deviations in the desired forces, pure position control is inappropriate for tasks that require physical contact with the environment. The proposed algorithm takes as input the reference trajectory and force profile, then adapts the robot movement by introducing small offsets to the reference trajectory so that the resulting forces exerted by the patient match the reference profile. The learning procedure is based on Dynamic Movement Primitives (DMPs).