Reproduce “WorldBridge: Bridging Video and Action for Task Generalization in World Action 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
Jun Cen, Yubo Guo, Hao Li, Liangkai Li, Yaoxu Lyu, Meng Wei, Zeqi Xiao, Sizhe Yang and 4 more
WorldBridge predicts future task videos and converts them into robot actions through jointly trained video and action experts. On a fixed RoboTwin 2.0 out-of-distribution task partition, average success rises from 25.07% with in-distribution supervision to 55.90% with unlabeled task videos and 70.70% with one labeled demonstration per task. These results show complementary benefits within this experiment.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.