Reproduce “Learning Self-Orchestration in Decentralized Agent Swarms” 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
Guowei Xu, Zhenting Qi, Amani Kiruga, Himabindu Lakkaraju, Yilun Du
MPT lets decentralized agents bid for turns, learning both actions and participation decisions from task feedback. The winning bid sets how credit flows to actors. On held-out Minecraft crafting seeds, MPT achieves mean episode reward of 41.5 ± 5.9, versus 20.3 ± 4.1 for monolithic RL. Rewards are task-specific and capped at 140; MPT relies on external evaluation and counterfactual rollouts.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.