Imitation Learning Control Strategy Based on Probabilistic Motor primitive for Upper Limb Rehabilitation Training

This paper proposes an imitation learning control strategy based on Probabilistic Movement Primitives (ProMP) for upper limb rehabilitation training. The strategy involves employing a probabilistic model to capture the probability distribution of motion data, enabling the planning of rehabilitation trajectories that align with the patient’s movement patterns, address limitations in motion amplitude and range, and account for uncertainty and variability in motion. This approach aims to enhance robustness in the face of noise and interference. Motion trajectory data of participants were collected using the Vicon motion capture system, and compared using the ProMP algorithm with traditional Dynamic Movement Primitives (DMP). The results indicate that trajectories generated by ProMP are smoother, more consistent, and outperform DMP in evaluation criteria such as smoothness of motion, velocity standard deviation, acceleration standard deviation, and curvature standard deviation. Furthermore, the paper discusses the utilization of speed heat maps in rehabilitation training. Speed heat maps enable the identification of high-pressure areas during movement, facilitating the optimization of rehabilitation training strategies, risk reduction, and improvement in rehabilitation outcomes. The optimized rehabilitation trajectory serves as a foundation for personalized rehabilitation plans, presenting innovative approaches to enhancing the efficacy of rehabilitation training.