Reproduce “Bayesian Offline Multi-Agent Reinforcement Learning with Latent World Models” 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
Xi Xiong, Lu Liu, Yiming Li, Jakob Foerster
The authors propose MPIL, an offline method for refining models and training policies in partially observable multi-agent settings. It summarizes agents’ histories in a latent state and uses an ensemble reliability signal to refine models offline. On Walker2d full-replay, it reports normalized average return of 63.57 ± 11.93, versus 26.52 ± 7.28 for MOMA-PPO; evidence is limited to three MuJoCo tasks and coverage assumptions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.