Real time gait generation for autonomous humanoid robots: a case study for walking

摘要

As autonomous humanoid robots assume more important roles in everyday life, they are expected to perform many different tasks and quickly adapt to unknown environments. Therefore, humanoid robots must generate quickly the appropriate gait based on information received from visual system. In this work, we present a new method for real time gait generation during walking based on Neural Networks. The minimum consumed energy gaits similar with human motion, are used to teach the Neural Network. After supervised learning, the Neural Network can quickly generate the humanoid robot gait. Simulation and experimental results utilizing the "Bonten-Marn I" humanoid robot show good performance of the proposed method.