Reinforcement Learning for Multi-Agent Planning in the League of Robot Runners

Alexandra Barancikova, Reuth Mirsky

Keywords: Multi-Agent Planning; Reinforcement Learning; Monte Carlo Tree Search; Robot Coordination

TL;DR: A hybrid reinforcement learning and planning approach that adapts Monte Carlo Tree Search for multi-agent coordination in the League of Robot Runners.

Abstract: Integrating planning and learning remains a central challenge in developing adaptive multi-agent systems. This work investigates how reinforcement learning (RL) can enhance multi-agent planning through experiments in the League of Robot Runners (LoRR), a benchmark traditionally rooted in classical planning. We reformulate LoRR in a multi-agent RL environment and adapt the Monte Carlo Tree Search (MCTS) algorithm for use in these settings. Because standard MCTS assumes a single agent and does not reuse experience, our approach introduces limited information sharing across episodes to improve coordination and efficiency. Preliminary results show that while the adapted method does not yet match the performance of classical planners, it provides a foundation for exploring how learning-based techniques can enhance coordination and adaptability in structured planning environments.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 14