Hierarchical Reinforcement Learning for Swarm Confrontation with High Uncertainty

• DOI: 10.48550/arxiv.2406.07877
• Publication Date: 2024-06-12
• Authors: Qizhen Wu; Kexin Liu; Lei Chen; Jinhu Lü
• Keywords: FOS: Computer and information sciences; Computer Science - Robotics; Computer Science - Machine Learning; Artificial Intelligence (cs.AI); Computer Science - Artificial Intelligence; Robotics (cs.RO); Machine Learning (cs.LG)

In swarm robotics, confrontation including the pursuit-evasion game is a key scenario. High uncertainty caused by unknown opponents' strategies and dynamic obstacles complicates action space into hybrid decision process. Although deep reinforcement learning method significant for since it can handle various sizes, as an end-to-end implementation, cannot deal with Here, we propose novel hierarchical approach consisting of target allocation layer, path planning underlying interaction mechanism between two layers, which indicates quantified uncertainty. It decouples process discrete continuous probabilistic ensemble model to quantify regulate frequency adaptively. Furthermore, overcome unstable training introduced design integration pre-training cross-training, enhances efficiency stability. Experiment results in both comparison ablation studies validate effectiveness generalization performance our proposed approach.