Dynamic Simulation Framework of the Robot-Assisted Training Platform (RATP)

Abstract

This paper introduces an innovative approach that seamlessly integrates a Genetic Algorithm (GA)-based method for generating and optimizing gait patterns with inverse dynamics analysis within the realm of personalized robot-assisted rehabilitation/training. Unlike conventional methods reliant on experimental data, our proposed approach enables the estimation and prediction of essential biomechanical information, including joint loads (joint force and torque). It achieves this by harnessing data derived from a customized musculoskeletal model for each user. By incorporating Lagrangian dynamics into the same platform used for gait pattern generation, the obtained gait pattern serves as a direct reference value for joint torque in applications such as robotic rehabilitation/training using wearable exoskeleton robots. Even in medical scenarios where only gait patterns are available using biomechanical information, our method can effectively estimate these parameters. We validated our method using a motion capture (MoCap) dataset as ground truth. This comparison assessed how well our estimations of joint angles, torques, and ground reaction forces (GRFs) matched actual human movement data. The results demonstrated high similarity, with an average Pearson Correlation Coefficient (PCC) value of 0.986 for joint angles, 0.748 for joint torques, and 0.975 for GRFs.

Keywords

• Legged locomotion
• gait recognition
• Lagrangian functions
• trajectory optimization
• motion planning
• robot kinematics