Reproduce “Video Prediction Policy 2: Predict Better, Act Better” 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
Yanjiang Guo, Haodong Yan, Zhide Zhong, Zhongru Zhang, Qingyuan Yang, Qingzhou Lu, Xiaoyu Chen, Yen-Jen Wang and 10 more
VPP2 predicts a described manipulation event from video, then uses that forecast to generate short action sequences. It is trained first on complete events and then on fixed-duration video chunks aligned with action learning. In real-world ALOHA tests without task-specific fine-tuning, it averages 58.5% success across 10 categories, versus 40.0% for π0.5; ALOHA was seen during training.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.