Computing and Informatics

Multi-agent collaborative path planning focuses on how the agents have to coordinate their displacements in the environment to achieve different targets or to cover a specific zone in a minimum of time. Reinforcement learning is often used to control the agents' trajectories in the case of static or dynamic targets. In this paper, we propose a multi-agent collaborative path planning based on reinforcement learning and leader-follower principles. The main objectives of this work are the development of an applicable motion planning in a partially observable environment, and also, to improve the agents' cooperation level during the tasks' execution via the creation of a dynamic hierarchy in the pursuit groups. This dynamic hierarchy is reflected by the possibility of reattributing the roles of Leaders and Followers at each iteration in the case of mobile agents to decrease the task's execution time. The proposed approach is applied to the Multi-Pursuer Multi-Evader game in comparison with recently proposed path planning algorithms dealing with the same problem. The simulation results reflect how this approach improves the pursuit capturing time and the payoff acquisition during the pursuit.