Optimizing Human Gait by Reducing Metabolic Cost with Imitation Learning

This study presents a hybrid imitation learning (IL) approach to optimize human gait simulation by minimizing metabolic cost using only kinematic data. The proposed method integrates a biomechanically informed reward function into the Generative Adversarial Imitation Learning (GAIL) framework, combining kinematic imitation with metabolic efficiency. Using the LocoMuJoCo platform and an OpenSim-based musculoskeletal model, the Hybrid IL approach demonstrated improved muscle activation patterns, reduced metabolic costs, and enhanced ground reaction force (GRF) accuracy compared to traditional methods. The findings highlight the method's potential for developing advanced assistive devices, including exosuits, exoskeletons, and prosthetics, offering effective solutions for individuals with gait impairments.