GroundingPI：面向物理智能的视觉基元 grounding 基础模型

AI 摘要

研究团队提出 GroundingPI，一个 4B 参数的 grounding 基础模型，将点和框以量化坐标形式生成于共享词表中，训练结合多模态与空间预训练、监督微调及 GRPO 强化学习。该模型在 34 个 grounding 基准、11 项感知能力上平均达 73.68%，超过更大的 GPT-6 Astra（71.54%），并在 RoboTwin 2.0、RoboCasa-GR1 和 nuScenes 等机器人操作与自动驾驶下游任务中提升表现。作者认为 grounding 可作为感知基础，专用感知预训练是通向物理智能基础模型的有前景方向。

正文节选

GroundingPI: A Grounding Foundation Model towards Physical Intelligence with Visual Primitives Abstract Precise grounding matters. It specifies which object is the target and where that object is, even in clutter and for tiny objects, and it has to be fast enough for closed-loop control. Yet vision-language-action (VLA) and world-action models (WAMs) take perception from general-purpose vision-language and video-generation backbones, which still fail in these settings. We introduce GroundingPI,

发布时间：—

抓取时间：2026-10-13 13:22

来源机构：Hugging Face