Learning Gait Control for Quadrupedal Robots Operating on Ships

To deploy quadruped robots for inspection tasks on ships, the locomotion controller must maintain stability under the six degrees-of-freedom motion of the ship. In this work, we propose a robust locomotion control policy trained using the proximal policy optimization algorithm to maintain robustness against dynamic disturbances from ship motion. To achieve this, a ship motion simulation platform is developed in Isaac Sim, where periodic roll and pitch motions are applied to simulate realistic ship dynamics. In addition, a reward function to improve gait stability and velocity tracking under dynamic disturbances is developed, and the training hyperparameters are tuned to enhance robustness. The proposed controller is compared with controllers trained on flat and unstructured rough terrain. The simulation results show that all three policies stably converged, achieving stable walking on the terrains used for training each controller. However, under disturbance conditions such as ship motion, the proposed policy maintained a stable posture and reliably tracked the commanded speed, whereas the baseline controllers failed to maintain a stable performance. These results demonstrate the effectiveness of the proposed controller for onboard inspection tasks.