Cooperative planning of multi-UAV logistics delivery by multi-graph reinforcement learning

Abstract

In this paper, we design a deep reinforcement learning algorithm based on graph neural network to solve the problem of cooperation control of multiple UAVs. Our algorithm can control multiple UAV swarms to complete package delivery tasks in an unexplored area under partial observation. Since each UAV has only a limited observation space and a small range of communication, we propose a reinforcement learning algorithm based on graph neural network, which can process multiple graphs simultaneously time and aggregate the feature vectors of the neighbor agents to address the cooperative issue of heterogeneous multi-agent coordination. We conduct a couple of ablation experiments to prove the effectiveness and performance characteristics of our algorithm.