Deterministic Policy Gradient Based Formation Control for Multi-Agent Systems

This paper studies the problem of formation control of multi-agent systems with the reinforcement learning method. A novel multi-agent formation control algorithm is first proposed, which adopts the framework of centralized training with decentralized execution, and combines the deterministic policy gradient (DPG) method with multi-agent advantage function. Then, three scenarios under partial observable Markov games are presented to study the multi-agent formation control problem and verify the proposed algorithm. Simulation results show that the proposed algorithm is effective in achieving the multi-agent formation control tasks.