Reproduce “OnlineWM: Causality-Aware Active Online Learning for Effective World Modeling” 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

Yikun Miao, Fangqi Zhu, Quanxin Shou, Xiaoyi Pang, Zhengyang Yan, Junhao Li, Haodong Wang, Zicong Hong and 1 more

OnlineWM improves action-conditioned video prediction by using model errors to select simulator experiences and contrasting outcomes from the same starting state. Its online learning balances scene and action difficulty, while counterfactual fine-tuning distinguishes a selected action outcome from a visually similar alternative. On held-out Minecraft sequences, Combined accuracy reaches 0.425 versus 0.356 for Random, with transfer evidence limited beyond Minecraft.

Summary by alphaXiv.

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