Enhancing Amphibious Flipper‐Based Locomotion Through Terrain‐Dependent Bayesian Optimization of Central Pattern Generator Control

Sea turtles use flippers to locomote across aquatic and terrestrial environments despite substantially different mechanical demands. This multimodal capability motivates amphibious robots that can exploit flipper‐based propulsion across heterogeneous terrains, but gait performance depends strongly on environmental conditions. Here, we investigate terrain‐dependent flipper locomotion using a sea turtle–inspired robot controlled by a Hopf central pattern generator (H‐CPG). The controller parametrizes gait timing, oscillator amplitude, coupling, and stroke amplitudes for the foreflippers, hindflippers, and hip joints. We first perform simulation‐based sensitivity analysis to determine how CPG parameters affect forward speed and cost of transport across multiple environments using diagonal and synchronous coordination. We then employ multiobjective Bayesian optimization to identify terrain‐ and gait‐specific controllers that balance speed and energetic cost. Optimized controllers are transferred to hardware through a simulation‐guided, hardware‐calibrated workflow and evaluated across rigid, sandy, rocky, inclined, and aquatic testbeds. Hardware experiments show that diagonal coordination improves forward progression and electrical cost of transport across terrestrial, granular, rocky, and inclined terrains, whereas synchronous coordination performs better in water. These results demonstrate that effective amphibious locomotion requires terrain‐dependent gait selection and parameter tuning rather than a single fixed coordination pattern, providing design insights for bioinspired amphibious robots operating in heterogeneous environments.