Reproduce “FutureWorld: A Plug-and-Play Causal Next-State Predictor for Interactive Video World Exploration” 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
Guangyuan Li, Kelu Yao, Shuolin Xu, Xinlei Yu, Wei Xing, Lei Zhao, Peng-Tao Jiang, Xiaogang Xu and 1 more
FutureWorld adds an action-conditioned predictor to interactive video generators to guide the next scene. It forecasts the next latent chunk from the generated latent and upcoming action, then modulates the generator’s latent; completed output remains history. Across four augmented generators, benchmark metrics improve on three 60-second datasets. On RealEstate10K-test, Matrix-Game 3.0’s VBench average rises from 0.8124 to 0.8314, without guaranteeing rollout error contraction.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.