A deep learning based human gait Trajectory generator for bipedal robots

Abstract

Bipedal robots have gained considerable attention due to their diverse applications in newlinetransportation, manufacturing, rehabilitation, etc. However, replicating human-like newlinewalking in bipedal robots remains challenging, as human walking is a complex learning newlineprocess demanding the seamless coordination of the brain s motor cells with its leg newlinemuscles. Furthermore, ensuring walking stability in bipedal robots remains an area that newlinerequires further exploration and development. The traditionally developed newlinecomputational models for generating lower limb joint trajectories require a rigorous newlinemathematical formulation suited to the available data. As a result, these models cannot newlinebe generalized to accommodate varying body parameters. Recently, machine learning newline(ML) and deep learning (DL) algorithms have demonstrated reasonable efficacy in newlinegenerating human gait trajectories. However, these techniques have some operational newlinelimitations, like manual mathematical modeling for feature extraction, the need for newlineexpensive motion capture systems, and randomness in hyperparameter selection during newlineneural network training newline