An Investigation on Multi-UAVs Cooperative Control Algorithm for Target Chasing

Abstract

Aiming at the problem of collective pursuit of Multiple Unmanned Aerial Vehicles (multi-UAVs) against non-cooperative UAVs, a directional chase strategy based on the Multi-Agent Deep Deterministic Policy Gradient(MADDPG) algorithm is designed by using deep reinforcement learning theory. By designing the algorithm model, state variables, actoin variables and reward function, the UAVs are trained to learn the directional chase strategy on a distributed Actor, centralized Critic structure. It is verified by simulation that the proposed chase strategy has higher learning efficiency while ensuring accuracy than the strategy with distributed training using the Deep Deterministic Policy Gradient (DDPG) algorithm. It also has a higher capture efficiency than the conventional pursuit-only strategy, provides a new research idea for multi- UAV confrontation.