Multi-Bionic Fish Collaborative Navigation Simulation Based on Reinforcement Learning

In marine exploration and marine ecosystem simulation tasks, collaborative navigation of multi-bionic fish has always been a critical topic, and designing efficient navigation algorithms for it has long attracted significant attention from scholars in computer science, marine ecology, and other fields. However, marine environments are mostly harsh, and directly designing multi-bionic fish systems and conducting underwater experiments without verification by any simulation system face high risks and costs. Therefore, the vast majority of experiments must be simulated in a simulation system before implementation. To address this issue, this paper propose brand new system for multi-bionic fish collaborative navigation based on the Unity3D simulation environment. First, the convenient modeling tools of the Unity3D simulation environment are used to design a realistic underwater environment, providing a setting for multi-bionic fish collaborative navigation tasks. Then, a classic fish collaborative model Boid is introduced briefly. Finally, BMA is proposed, a combination between Boid model and a brand-new reinforcement learning-based navigation method. Simulation results show that the system not only well reproduces the ecological behaviors of real multi-bionic fish but also efficiently completes navigation tasks, with all indicators outperforming the built-in navigation methods of the Unity3D simulation platform.