关键词: 目标围捕, 强化学习, 势能函数, 多智能体系统, 避障

Abstract:

A target round-up control method for multi-agent systems is proposed based on reinforcement learning. Firstly, Markov game modeling for multi-agent systems is carried out. The potential energy function which meets the requirements of arriving at the desired state and avoiding obstacles is designed according to the task of rounding up, and reinforcement learning principles are combined with the model control. The round-up is performed using multi-agent reinforcement learning guided by the potential energy model. Secondly, based on the existing potential energy model, two surrounding strategies are established: tracking round-up and circumnavigation round-up. With the first strategy, consistent tracking of multiple agents is achieved by designing the potential energy function of velocity. In the second strategy, virtual circumnavigation points are added to design potential energy functions, achieving desired circumnavigation. Finally, the effectiveness of the round-up control based on multi-agent reinforcement learning is verified by simulation.

Key words: target round-up, reinforcement learning, potential energy function, multi-agent systems, avoiding obstacle