Cooperative Obstacle Avoidance of Unmanned System Swarm via Reinforcement Learning Under Unknown Environments

The deployment of unmanned system swarm, especially unmanned ground vehicles (UGVs) and unmanned aerial vehicles (UAVs), has gained significant attention in various applications. However, the effective navigation of unmanned system swarm in unknown and dynamic environments remains a formidable task. To address this challenge, a novel approach that combines reinforcement learning and artificial potential field (APF) methods is presented in this article. The proposed framework utilizes the double deep Q-network (DDQN) with a prioritized experience replay mechanism to train the actions of the virtual leader, which leads to the motion of the whole unmanned system swarm. In addition, the APF method is also applied to each unmanned system in swarm to achieve distributed obstacle avoidance. By training virtual leader to guide all individuals, the unmanned system swarm can adapt and navigate safely in unfamiliar surroundings. Simulations and experiments validate the proposed method, demonstrating enhanced obstacle avoidance and autonomous navigation for the swarm.