MADDPG: Multi-agent Deep Deterministic Policy Gradient Algorithm for Formation Elliptical Encirclement and Collision Avoidance

This paper considers the formation encirclement problem of multi-agent systems with collision avoidance, where the enclosing orbit is an ellipse, the target and the formation are dynamic. The novel rewards, that is the elliptical encirclement reward, the formation reward, the angular velocity reward and collision avoidance reward are designed and a reinforcement learning (RL) algorithm, that is multi-agent deep deterministic policy gradient (MADDPG), is designed based on the novel setting of rewards. Simulation results are given to show the validity of the proposed method.