Reproduce “Attention-Guided Contrastive Role Representations for Multi-Agent Reinforcement Learning” 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
Zican Hu, Zongzhang Zhang, Huaxiong Li, Chunlin Chen, Hongyu Ding, Zhi Wang
ACORM learns agent roles from behavior histories and uses the global state to attend to those roles when valuing coordinated actions. It combines contrastive learning with clustering and state-conditioned attention within centralized training and decentralized execution. On six StarCraft Multi-Agent Challenge maps, it performs best on all super-hard maps and most others; it still requires explicit clustering and a prespecified number of clusters.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.