Reproduce “ScaleVLA: Scaling Vision-Language-Action Models to 552 Billion Parameters” 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

Ruixi Yu, Dairu Liu, Xuchuan Chen, Zekun Qi, Zekai Li, Zhuo Cao, Luyang Niu, Yingsheng Zhang and 4 more

ScaleVLA connects a frozen 552-billion-parameter vision-language backbone to a trainable action head, which converts image, instruction, and robot-state features into action chunks without generating text. In fixed-budget benchmarks, it reaches 87.8% LIBERO-Goal success with 10 demonstrations per task, versus 85.6% for a 2B policy with 50. Its comparisons do not isolate parameter count from model-family differences.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.