Intelligent Control of Collective Motion Based on Deep Learning

According to the collective motion parameters of the red nose fish, a multi-parameters input deep neural network model is designed for individual hierarchical interaction. The model is trained by using the collective motion data of the real red nose fish. A vision-based key neighbor selection method is proposed, which is used for integrating the information of all neighbors. Then, with respect to the selected key neighbor, the deep neural network model of hierarchical interaction is applied. This method is verified by a self-made simulation software. The simulation result shows that the multi-agents group can perform self-organization collective motion behavior. Compared with other collective motion methods, the macroscopic properties of the simulation shows that our proposed intelligent control method is more consistent with the collective motion performance of the real fish.