Reproduce “Perception into Imagination: Executable Memory for 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

Xianyao Li, Ruitong Tian, Rui Min, Fang Xu, Jing Du

The authors introduce executable perceptual memory (EPM), which advances an explicit perceptual state with each candidate action and supplies its changing read-out to reward, value, and policy learning in imagined futures. On Memory Maze 13×13, EPM reaches 30.37 targets versus 23.37 for Evolved at 1M interactions. This full-method result does not isolate reward access, whose MiniWorld effect remains unresolved.

Summary by alphaXiv.

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