Reproduce “MemWM: Associative Memory-Enhanced Latent World Models for Long-Horizon Planning” 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

Yuwei Sun, Siyu Zhu

MemWM adds learned, content-addressable memory to a latent world model for visual planning. It retrieves a weighted blend of learned slots from an observation and conditions a shared predictor on that context before the immediate action. In Reacher, success is 90% at 25-timestep separation and 88.0% at 50; this result is specific to the tested setting.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.