VLA 模型全生命周期综述:覆盖 88 篇工作、分析 22 个模型
Vision-Language-Action Models: A Life-Cycle Survey of Design, Training and Deployment
做机器人方向的人值得收藏这份综述,88 篇论文按设计、训练、部署三条线串起来,22 个模型的数据规模和推理速度都有对比表。 π0、GR00T N1.7、SmolVLA 用什么数据训练一查就有,选型时省不少翻论文的时间。
一篇发布于 arXiv 的综述系统梳理了视觉-语言-动作(VLA)模型的发展,覆盖 2022 年 12 月至 2026 年 10 月间的 88 篇工作,从首个 VLA 模型 RT-2 讲到可在单张消费级 GPU 上微调的开源模型。综述沿七条轴线展开:输入与记忆、架构、动作输出、中间预测、预训练数据、后训练、实时部署,并对 22 个模型逐一分析其训练数据、参数量、推理速度和硬件需求。
数据来源的演变是核心观察之一:π0 使用约 10,000 小时遥操作机器人数据,SmolVLA 依赖 23,000 条社区采集片段,GR00T N1.7 则利用 20,000 小时第一人称人类视频。模型规模出现分化:多数新模型集中在 3-70 亿参数,另有一批低于 10 亿参数的模型可跑在消费级 GPU 上。UniVLA 和 Qwen-VLA 等通才型 VLA 开始同时处理操作与导航任务。
综述还指出一个关键问题:部分 VLA 在基准测试中得分接近满分,但物体或相机位置稍有移动就会失败,这揭示了当前评测体系的局限。对从事机器人学习的研究者和工程师来说,这篇综述提供了按生命周期组织的设计选型参考,并与世界-动作模型做了对比分析。
Vision-Language-Action Models: A Life-Cycle Survey of Design, Training and Deployment
Vision-language-action (VLA) models adapt web-pretrained vision-language models to output robot actions, so that web knowledge carries over to physical control. In under 4 years the field has moved from RT-2, the first VLA, to open models that fine-tune on a single consumer GPU and to proprietary generalists that clean unseen homes. This survey reviews 88 works from December 2022 to October 2026, mainly on manipulation and navigation, plus humanoid control, world-action models and evaluation. We follow a VLA through its life cycle, from model design to training and deployment, along seven axes. Model design covers (a) inputs and memory, what the model observes; (b) architecture, how its pipeline is built; (c) action output, how it produces motor commands; and (d) intermediate predictions, what else it predicts, from reasoning steps to future observations. Training covers (e) pretraining data, the mix of robot, web, human and simulated data, and (f) post-training, adaptation to the target robot. Deployment covers (g) real-time execution, keeping pace with the robot. We analyse 22 models on these axes and report their training data, model size, inference speed and hardware. Three trends emerge: (i) training data has broadened, from about 10,000 hours of teleoperated robot data ($π_0$) to 23,000 community-collected episodes (SmolVLA) and 20,000 hours of egocentric human video (GR00T N1.7); (ii) model sizes have split: most recent models have 3-7 billion parameters, while a new group under one billion runs on consumer GPUs; and (iii) generalist VLAs are emerging that handle both manipulation and navigation, such as UniVLA and Qwen-VLA. We also compare VLAs with world-action models, which predict future observations as well as actions. We review evidence that some VLAs with near-perfect benchmark scores fail when objects or the camera move slightly, and open up further research questions.
本条情报归属 AITOP「论文」核心追踪栏目。在跨区域产业观察坐标中,该动态直接关联「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:VLA、π0、GR00T N1.7、SmolVLA,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:本文原始线索采编自「arXiv cs.LG」(@Suvarna Kadam, Soran Parsa),经 AITOP 多维评分管线结构化提炼。