Learning and Control of Terrestrial Gait for a Bionic Turtle

This article applies deep reinforcement learning (DRL) to train and control the gait of a bionic turtle. A simulation environment is built on the IsaacGym platform to evaluate the terrestrial gait strategy of the bionic turtle. During training, the robot achieves stable tracking of various velocity commands and responds quickly to command switches while maintaining body stability throughout the transition. The trained gait exhibits strong adaptability and robustness, as evidenced by a stable gait phase diagram. Simulation validation confirms that, at a control frequency of 50Hz, the robotic turtle maintains body stability, avoiding excessive shaking or instability. This study demonstrates the significant potential and effectiveness of deep reinforcement learning in bionic turtle gait control, offering valuable insights for future research on bio-inspired gait control.