Reproduce “A Theoretical Framework for Parallel Lifelong MAPF Using Group Decentralized Planning” 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
Alex DeWeese, Jiaoyang Li, Guannan Qu
Researchers at Carnegie Mellon University developed a theoretical framework for Lifelong Multi-Agent Path Finding (L-MAPF) by modeling it as a Locally Interdependent Multi-Agent MDP, proving the near-optimality of the Rolling-Horizon Collision Resolution (RHCR) method and introducing Group Decentralized RHCR (GD-RHCR). GD-RHCR significantly reduces planning time by up to 24.9x and achieves up to 57.7% higher throughput compared to existing methods in high-agent-count scenarios.
Key insights
- Rolling-Horizon Collision Resolution (RHCR) is near-optimal in a discounted MDP L-MAPF formulation, with its suboptimality exponentially decreasing with the planning horizon (H) and unaffected by the replan window (α).
- Group decentralization, derived from Locally Interdependent Multi-Agent MDP (LI-MDP) theory, enables parallel planning with theoretical near-optimality guarantees linked to both temporal horizon and spatial visibility.
- Dynamic grouping, lazy evaluation, and a proactive secondary solver in GD-RHCR are crucial for reducing computational cost and gracefully handling high-agent counts and congestion.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.