Energy Efficient Swimming: Exploring an Intermittent Swimming Gait for Robotic Fish via Deep Reinforcement Learning

Abstract

Energy conservation is a major challenge for robotic fish due to difficulties in replenishing energy underwater. In response, we explore the optimal energy efficient swimming gait for robotic fish using our proposed deep reinforcement learning (DRL) framework. Surprisingly, our work reveals that traditional continuous swimming gaits are not the most energy efficient option for robotic fish. Instead, “smart swimmers” utilize intermittent swimming gaits (ISG) to enhance efficiency, resembling the natural “burst and coast” behavior observed in real fish. During the burst phase, the robotic fish utilizes high-frequency undulations to accelerate, followed by a coast phase where it remains stationary, analogous to surfing. The key difference in intermittent swimming at different speeds is the coast phase duration. Furthermore, our computational fluid dynamics analysis reveals the energy-saving mechanism behind this ISG for robotic fish. Experimental results confirm the superiority of the ISG learned from DRL over the commonly used continuous swimming gait in robotic fish applications, resulting in a notable average energy savings of 24.2% .