A Teleoperation Framework for Upper Limb Rehabilitation Based on Improved Dynamic Movement Primitives and Hybrid Control

To achieve efficient rehabilitation training for hemi-plegic limbs of stroke patients conducted by therapists in the tele-operation environment, we propose a robot skill learning method based on improved dynamic movement primitives (DMPs) and virtual fixture (VF) based on a hybrid control strategy in this paper. To overcome the noise problems faced by traditional DMPs when learning from a single trajectory, Gaussian Mixture Model - Gaussian Mixture Regression (GMM-GMR) is introduced. This approach extracts the main movement features from multiple demonstration trajectories, facilitating autonomous training for the patient without the real-time guidance of a therapist. In addition, the hybrid control method utilizes the muscle activation obtained from the surface electromyography signal processing. It incorporates variable stiffness control reflecting the patient's control intention on top of position error-based proportional-derivative control, thus helping patients train interactively under therapist-driven training. Finally, the experiments conducted with a Sawyer robot and a Touch device have confirmed the effectiveness of the proposed methods.