Physical Intelligence's π Series: A Universal Robot Brain
0营收，估值200亿美元，这家公司想成为机器人的安卓
Physical Intelligence, a U.S. company with zero revenue but a $20 billion valuation, is developing a universal 'brain' for robots through its π series of models. These models aim to provide cross-robot and cross-task capabilities, with recent versions like π0.7 enabling steering and new tasks. The company's valuation is based on its potential to become the 'Android of robotics,' but its lack of revenue and the uncertainty around the convergence of robot intelligence remain key challenges.
Source: Chinese robotics — Deployment discovery · Read original article ↗
English translation unavailable. Choose the original article to read it.
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
训练数据是8种不同机器人、1万多小时的遥操作。
Open source S15
2025年4月，π0.5，这一代解决的是泛化，把一个移动机器人放进完全没见过的人家里，让它收拾厨房、整理卧室，一次任务10到15分钟。
Open source S21
这不是论文堆出来的漂亮曲线，Ultra是美国的一家仓储打包公司，用π0.6跑完了完整的一个班次，96.4%的自主率。
Open source S23
它卖的不是“机器人会干活”，卖的是同一套大脑，能装进任何机器人。
Open source S24
2024年10月，其开发的具身智能基础模型π0，用PaliGemma做视觉语言底座，外挂一个33亿参数的“动作专家”，靠 flow matching 直接输出50赫兹的连续动作。
Open source S14
靠的不是堆更多机器人数据，而是 异构共训练 ——网络图文、人类口头指导、子任务标注、其他机器人的数据，全揉进一个训练目标里。
Open source S22
PI自己不做机器人，直接买3500美元一台的现成机械臂，在办公室、真实家庭、仓库、商用测试厨房里采数据。
Open source S19
Source：Chinese robotics — Deployment discovery · news.qq.com