Reproduce “Arm-wise Compositional Generalization in Dual-Arm Vision-Language-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
Zaibin Zhang, Binghao Ran, Yuhan Wu, Zhongbo Zhang, Yifan Wang, Junwei Jiang, Junlan Xiao, Wangcheng Shi and 5 more
AE-VLA structures a shared policy to reuse familiar arm-level skills under changed coordination. It groups action tokens by arm, adds skill-specific adapters, and restricts direct cross-arm action attention while retaining shared scene context. On held-out simulation conditions, it achieves 21.53% constraint-compliant success versus 5.53% for Dual π₀.₅; on physical robots, it averages 39.00% across five unseen conditions. Tests use supplied prompts, and synchronization remains difficult.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.