Frontier vision language models improve robot learning and control

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

Robotics

Summary

Big vision-language models, like those used to understand pictures and text, might help robots learn better how to act in the real world. The researchers connected a huge model with 552 billion parameters to a robot action system and found it learned tasks faster and performed more accurately than smaller models. Tasks included things like precise object alignment and general robot tasks. Even with its large size, the system responded fast enough for practical robot control. This work suggests that very large vision-language models can be a solid base for teaching robots to do complex actions.

What this means in practice

  • For robotics engineers: Improve robot task learning speed and accuracy by using very large vision-language models as the foundational understanding system.
  • For industrial automation teams: Enhance precision in contact-sensitive robot tasks such as electronic device assembly by integrating large vision-language models with action modules.

Authors

Ruixi Yu, Dairu Liu, Xuchuan Chen, Zekun Qi, Zekai Li, Zhuo Cao, Luyang Niu, Yingsheng Zhang, Sikai Liang, Ke Fan, He Wang, Li Yi

Abstract

Frontier 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.