Random Latent Exploration-Based Multi-Agent Reinforcement Learning for Multi-UAV Path Planning

Unmanned aerial vehicles (UAVs) have become increasingly popular in civilian and military applications due to their low cost and high maneuverability. Multi-UAV path planning is a highly challenging problem with an NP-hard complexity, and frequently solved by machine learning algorithms, especially in dynamic and open environments. However, the sparse-reward problem in traditional machine learning algorithms often makes solutions trapped in local optima and results in unsatisfactory f light paths, seriously affecting the coordination effect of UAVs. In this paper, we propose an enhanced random latent exploration based multi-agent reinforcement learning framework to provide a reasonable flight path for each UAV and efficiently achieve the group missions. First, we analyse the constraints and objectives of the multi-UAV path planning problem and abstract it as a multi-constraint combinatorial optimization one. Then, inspired by exploration bonus and randomized value function mechanisms, we propose a random latent exploration-based multi-agent reinforcement learning framework to enable UAVs to acquire more diverse rewards in path planning. Simulation experiments in a multi-agent particle environment are conducted to verify the effectiveness and efficiency of the proposed approach. Compared with baselines, our approach shows significant advantages and achieves a 71.2% increase in average reward, a 46.5% improvement in success rate, a 26% reduction in flight time, and a 33.6% reduction in path length.