A multi-UAV cooperative path planning method based on HCA-MAPPO deep reinforcement learning

To address the challenges of multi-UAV cooperative path planning, such as the flattening of perception information processing and low exploration efficiency under sparse rewards, this paper proposes a novel cooperative path planning method based on multi-agent deep reinforcement learning. This method models the multi-UAV cooperative path-planning problem as a partially observable Markov decision process and employs a deep reinforcement learning algorithm for solution. To enhance the agent’s ability to extract features and assess situational awareness from multi-source heterogeneous observation information, a hierarchical cross-attention-based agent network is designed, which achieves priority perception and processing of multi-source heterogeneous information through a sequential processing flow. Furthermore, an intrinsic curiosity module based on cooperative consistency gating is constructed to effectively couple individual exploration with cluster cooperative goals, improving the algorithm’s effective learning efficiency under sparse reward environments. Finally, simulation experiments verify the effectiveness and superiority of the proposed algorithm.