Reproduce “WorldProbe: When Video World Models Fall Apart under Long-Horizon Spatial 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
Minqi Wang, Jie Peng, Hejun Dong, Ruchang Yao, Puyi Wang, Siyuan Huang, Jihai Zhang, Yafu Li and 3 more
WorldProbe evaluates whether video world models maintain worlds during adaptive camera exploration. An agent chooses each next camera action from generated observations, and the benchmark measures quality, memory, and evolution across 10 models, 164 cases, and horizons up to 180 seconds. AlayaWorld scores highest on memory at 6.53/10, while DreamX-World leads evolution at 5.17/10, showing consistency remains limited in tests.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.