Reproduce “FineART: Fine-Grained Annotated Robotic Trajectory Dataset and Vision-Language-Action Model for Bimanual Manipulation” 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
Jade Choghari, Pepijn Kooijmans, Mansi Agarwal, Yusuf Umut Ciftci, Aseem Doriwala, Catherine Weaver, Mouli Sivapurapu, Kai Yang and 3 more
FineART pairs dense subtask labels across 40,543 episodes with a single policy that predicts the next language instruction and conditions bimanual actions on it. After ALOHA mid-training and fine-tuning on five YAM task types, it completed held-out prepare-breakfast sequences in 28% of rollouts, with 78.3% average progress; this shows transfer after YAM fine-tuning, not zero-shot transfer or reliable autonomous completion.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.