titleScaleVLA: Scaling Vision-Language-Action Models to 552 Billion Parameters
authorsRuixi Yu, Dairu Liu, Xuchuan Chen, Zekun Qi, Zekai Li, Zhuo Cao, Luyang Niu, Yingsheng Zhang, Sikai Liang, Ke Fan, He Wang, Li Yi
date2026-11-16
venuearXiv (preprint, not peer reviewed)
idarXiv:2611.21607
affiliationgalbot
upvotes36 (Hugging Face Daily Papers, 2026-11-19)
linksabs pdf hf
cite
primerVision-language-action models
relatedHumanoid-GPT: Scaling Data and Structure for Zero-Shot Motion Tracking
Where is Physical RSI capability from? An Observational Scaling Law of LLM-as-policy
Systematically Exploring the Capabilities of GPT-6 Astra as Embodied Policies
HumanTracker: Towards Comprehensive and Human-Aligned Motion Tracking Benchmark
LIMMT: Less is More for Motion Tracking
abstractFrontier Vision-Language Models (VLMs) such as GPT-6 Astra have demonstrated remarkable capabilities in robot control, suggesting that broad pretraining provides knowledge valuable for physical action. Yet most Vision-Language-Action (VLA) policies still rely on much smaller VLM backbones, leaving this potential largely untapped. We ask whether frontier-scale representations can provide a stronger foundation for robot learning. We introduce ScaleVLA, which connects a 552B DeepSeek-V4.1-Flash backbone to a trainable action DiT, and systematically compare VLMs from 2B to 552B under a shared training setup. We evaluate three fundamental capabilities of robot foundation models. In sample efficiency, just 10 demonstrations outperform a 2B backbone trained with 50 (87.8% vs. 85.6%). In generalization, LIBERO-PRO success reaches 45.1%, compared with at most 31.3% for smaller backbones. In precise manipulation, real-world charger insertion improves from 4/10 to 9/10 successes, with gains concentrated in contact-sensitive alignment. Despite its scale, ScaleVLA achieves 216 ms median latency per action chunk. Our findings suggest that frontier VLMs can serve not only as reasoning agents, but also as powerful representational foundations for efficient and generalizable robot policies.

bibtex

@misc{yu2026scalevla,
  title={ScaleVLA: Scaling Vision-Language-Action Models to 552 Billion Parameters},
  author={Ruixi Yu and Dairu Liu and Xuchuan Chen and Zekun Qi and Zekai Li and Zhuo Cao and Luyang Niu and Yingsheng Zhang and Sikai Liang and Ke Fan and He Wang and Li Yi},
  year={2026},
  eprint={2611.21607},
  archivePrefix={arXiv},
  url={https://arxiv.org/abs/2611.21607}
}