Employing machine learning for more efficient robotic fish

There are many fish in the sea, and most of them are very fast. To understand fish, it sometimes helps to try to imitate them, robotically. But thanks to millions of years of evolution, fish can achieve speeds and efficiency levels that today’s robotic facsimiles can only hope to match.

Cui et al. attempted to close that gap by employing deep reinforcement learning (DRL) to optimize the motion of a robotic fish. Their approach resulted in increased swimming efficiency and reduced energy usage.

Biomimetic robotic fish often attempt to emulate the flexible bodies of real fish through a segmented multi-link structure, but even a simplified model can struggle with the many parameters and changing conditions. DRL is well-suited to these types of problems, but even with this method, the authors had to take steps to reduce the complexity of the problem.

“In our paper, direct application of DRL did not yield optimal results due to several challenges, including insufficient agent perception of hydrodynamic environments and a vast action space for optimization,” said author Xinyu Cui. “To address these challenges, we employed several techniques to enhance the agent’s perception of hydrodynamic environments and reduce the optimization space for faster training.”

This innovative approach resulted in more efficient swimming dynamics than traditional computational methods. The authors are planning to expand their research by examining more complex models.

“Our next steps involve a more sophisticated exploration into the dynamic modeling of fish bodies to attain even greater precision,” said author Boai Sun. “We aim to integrate refined fluid-structure interaction models with advanced AI algorithms, targeting a holistic mastery over the flexible movements of fish bodies in various flow conditions.”

Source: “Enhancing efficiency and propulsion in bio-mimetic robotic fish through end-to-end deep reinforcement learning,” by Xinyu Cui, Boai Sun, Yi Zhu, Ning Yang, Haifeng Zhang, Weicheng Cui, Dixia Fan, and Jun Wang, Physics of Fluids (2024). The article can be accessed at https://doi.org/10.1063/5.0192993 .