A study on conditions for gait generation in a hexapod robot using deep reinforcement learning

In recent years, various approaches have been made in the control of robots using machine learning. Among them, research on deep reinforcement learning, which combines deep learning and reinforcement learning, has progressed. This learning method enables the learning of continuous behaviors by storing discretized Q-tables in neural networks. Examples of its application include tripod gait generation for quadruped robots and trajectory planning for industrial manipulators, contributing to learning complex behaviors. In this study, using virtual environments, we investigate reward conditions for hexapod robots performing tripod gaits, analyze variations under altered rewards, and identify key behavior factors．