HySTAR: Anchored Hypergraphs for Stable Credit Assignment in Cooperative Multi-Agent Reinforcement Learning
HySTAR anchors a fixed hypergraph so cooperative MARL assigns credit without structural target drift.
Cooperative multi-agent RL under partial observability must assign a shared reward to agents and coalitions. HySTAR, built on MAPPO, separates adaptive representation learning from a temporally consistent high-order value decomposition by anchoring an overlapping sparse hypergraph. A spatiotemporal encoder captures physical and task interactions, and temporal plus structural relevance produce agent-specific advantages. On SMAC, GRF, Traffic Junction, and MPE it outperforms MAPPO-style, value-factorization, and dynamic-grouping baselines, with relative gains of 16.7% over MAPPO and 15.6% over HYGMA on the hardest SMAC settings.
- Anchored hypergraph prevents structural target drift in credit assignment.
- MAPPO-based design uses a spatiotemporal interaction encoder.
- Hardest SMAC: 16.7% over MAPPO and 15.6% over HYGMA.
- Ranks first on all six GRF scenarios; up to 40.2% faster than MAGIC.
From the original article
1 min read at arxiv.orgCooperative multi-agent reinforcement learning under partial observability and shared rewards requires assigning team outcomes to individual agents and high-order coalitions. A MAPPO-style critic compresses joint behavior into one global value, while critics that dynamically reconstruct the grouping topology change the mapping from agents and coalitions to value components as interactions or active agents evolve.
By Xinglong Luo · arxiv.org
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.