论文多源确认精选

ScaleVLA:用 552B 参数 VLM 骨干做机器人策略,LIBERO-PRO 达 45.1%

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

精选理由

这篇论文把 552B 的 DeepSeek-V4.1-Flash 直接当机器人策略骨干用,10 条演示就打赢小模型 50 条的效果,LIBERO-PRO 也冲到 45.1%。做机器人或 VLA 方向的研究者值得对比自家 2B 骨干看看差距在哪。

论文 ScaleVLA 将 552B 参数的 DeepSeek-V4.1-Flash 作为视觉-语言-动作(VLA)模型的骨干,接上一个可训练的动作 DiT,并在 2B 到 552B 的统一训练设置下系统对比不同规模 VLM 的机器人控制表现。

正文

实验结果显示三项能力的提升:样本效率方面,仅用 10 条演示即达到 87.8% 成功率,超过 2B 骨干用 50 条演示的 85.6%;泛化方面,LIBERO-PRO 成功率达 45.1%,而小骨干最高为 31.3%;精细操作方面,真实插充电器任务从 4/10 提升到 9/10,增益集中在接触敏感的对齐环节。尽管规模巨大,其动作块中位延迟仍为 216 ms。

该工作论证了前沿级 VLM 可作为高效、可泛化机器人策略的表征基础,而不只是推理工具,对机器人基础模型的研究路线具有直接参考价值。

原文 · arXiv: DeepSeek

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

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.

产业坐标与信源核验

本条情报归属 AITOP「论文」核心栏目,重点聚焦中国 AI 产业与国产大模型生态。在跨区域产业观察坐标中,该动态直接关联「中美大模型技术代差与商业化对照」「AI 开发者工具链、Coding Agent 与工程化实践」研究主线,反映出当前技术演进与商业落地的最新风向。涉及核心实体与技术议题包括:ScaleVLA、DeepSeek-V4.1-Flash、VLA、机器人,读者可通过下方标签进入对应专题页查阅历史脉络与关联对比。

信源核验说明:本文原始信息发布自一手权威信源「arXiv: DeepSeek」(Tier-1 一手发布),具备高确定性参考价值。截至目前,全网已有 3 家独立信源对该事件进行交叉印证与跟进报道(含 pandaily、爱范儿 等),属于近期高关注度行业信号。