Reproduce “StarVLA-$α$: Reducing Complexity in Vision-Language-Action Systems” 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
Jinhui Ye, Ning Gao, Senqiao Yang, Jinliang Zheng, Zixuan Wang, Yuxin Chen, Pengguang Chen, Yilun Chen and 2 more
StarVLA-α tests whether a simple vision-language-action policy can serve as a strong baseline for robot tasks. A pretrained Qwen3-VL backbone and MLP action head convert camera images and instructions into continuous action chunks, without robot-state or history inputs. On 11 ARX5 RoboChallenge tasks, it achieves 33.6% success versus π₀.₅’s 12.7%, a 20.9-point gap, but this cross-paper comparison is not architecture-controlled.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.