Reproduce “Lucid Dreaming for World Models: Learning to Doubt Imagination and Decide by Trust” 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

Ziqi Wen, Ting Xu, Lianyu Wang, Xian Lin, Yanda Meng, Huazhu Fu, Meng Wang, Ching-Yu Cheng

LucidWM augments categorical world models with evidence-based doubt about predicted transitions and carries that doubt forward as trust during imagined rollouts. Its uncertainty readout uses existing transition-head logits, without added parameters or forward passes. In a DreamerV3 cheetah test, its alarm preceded all 20 falls, while prediction entropy alarmed on none; this result is limited to that setting.

Summary by alphaXiv.

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