Decentralized Master-Mind: Joint Action Refinement through Iterative Intent Denoising in Multi-Agent Pathfinding
DMM iteratively denoises joint action intents, solving 1,598 of 1,600 MovingAI pathfinding tasks.
Decentralized Master-Mind (DMM) replaces one-shot per-agent action sampling with discrete iterative refinement of action intents across communication rounds, inspired by diffusion denoising. Agents are pretrained by imitation learning on expert multi-agent pathfinding solutions and fine-tuned with MICPO, a critic-free group-relative reinforcement-learning method. On 1,600 MovingAI tasks, MICPO-tuned DMM solves 1,598 with costs close to the strongest baselines, and it scales beyond one million simultaneous agents in obstacle-rich maps.
- Independent per-agent sampling can form incompatible joint actions.
- DMM iteratively refines action intents over communication rounds.
- MICPO fine-tuning solves 1,598 of 1,600 MovingAI tasks.
- The method scales past one million simultaneous agents.
From the original article
1 min read at huggingface.co
Decentralized multi-agent path finding (MAPF) with communication requires agents to reach individual goals without collisions under partial observability. Learnable policies trained on expert data provide an effective approach to this problem.
By Valeriy Vyaltsev, Anton Andreychuk, Taisia Zlotnikova · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.