Reproduce “Alignment Drives Reuse: Learning Generalizable Robot-Use Agents from Heterogeneous Experience” 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
Ziyue Wang, Yaopeng Yang, Liyi Zhou
UniCraft makes robot experience reusable by aligning task cases, scene geometry, and robot actions, then using mixed-task rollouts to revise a shared skill repository and orchestration policy. In simulation, GPT-6 Astra reaches 89.25% full-task success in RoboTwin 2.0’s pooled unseen-scenario-and-unseen-task comparison, versus 10.00% for X-VLA. This comparison does not isolate alignment; transfer is not validated on physical robots.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.