Swarm Motion of Underwater Robots Based on Local Visual Perception

Many large schools of fish exhibit collective behaviors, such as large-scale migration and the formation of cylindrical arrays to evade predators. Previous studies have shown that these collective behaviors do not rely on global explicit information exchange but are achieved through visual observation of neighboring individuals. Given the unique characteristics of the underwater environment, a global positioning system similar to that on land has not yet been developed. Therefore, it is meaningful to investigate swarm motion based on local information exchange for underwater robots. Drawing inspiration from the swarm motion of fish, this study focuses on the swarm motion of underwater manta ray-like robots based on local visual perception information interaction. Firstly, a deep convolutional neural network method is employed to design a neighbor detection algorithm. This algorithm enables real-time acquisition of relative distance, orientation, and tracking ID information of neighboring robots. The proposed method is deployed on underwater manta ray-like robots, and a series of underwater experiments are conducted. The experimental results demonstrate improved detection accuracy and processing speed. Subsequently, a swarm motion model based on local visual perception is proposed. The neighbor detection information obtained in the experiments is utilized as constraint information for the simulation-based swarm motion model. The results indicate that swarm motion of the robot can be achieved through the acquisition of neighbor robot information. The research is based on local visual perception of group movement, which can make up for the inability to achieve global positioning in the underwater environment. And this research provides a framework for underwater swarm motion in the context of manta ray-like robots.