Reproduce “Generalizable Collaborative Search-and-Capture in Cluttered Environments via Path-Guided MAPPO and Directional Frontier Allocation” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Jialin Ying, Zhihao Li, Zicheng Dong, Guohua Wu, Yihuan Liao

PGF-MAPPO, a hierarchical framework for multi-robot collaborative search-and-capture, integrates path-guided Multi-Agent PPO with a Directional Frontier Allocation strategy. This approach achieves up to a 54% capture success rate against faster evaders in unseen 20x20m maps, demonstrating zero-shot generalization and improved exploration efficiency compared to existing methods.

Key insights

  • Integrating classical topological path planning (A*) with reinforcement learning effectively addresses sparse rewards and provides robust long-range guidance, crucial for generalization in cluttered environments.
  • Enforcing spatial diversity in exploration through a directional frontier allocation strategy prevents agents from clustering, significantly improving map coverage and search efficiency.
  • Parameter-shared, decentralized MARL policies, when coupled with high-level guidance, can achieve strong zero-shot generalization to much larger and more complex unseen environments.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.