Physical Intelligence
Physical Intelligence(Pi)正在打造用於機器人控制的基礎模型。其 pi0 模型是最早能在多種操作任務中控制各式機器人的視覺-語言-動作(VLA)模型之一。他們透過 openpi 框架開源模型,讓任何人只需 1-20 小時的任務資料,就能微調並部署機器人策略。
- 公司類型: company
- 商業模式: Foundation models for robot control
- 成立: 2024
- 總部: San Francisco, California, USA
- 員工人數: ~100
- 執行長: Karol Hausman
- 狀態: active
- 創辦人: Karol Hausman, Sergey Levine, Chelsea Finn, Brian Ichter
- 團隊規模: ~100 employees
- 重點領域: Foundation Models, Robot Learning, Manipulation, Vision-Language-Action Models
- 標籤: foundation-model, VLA, manipulation, open-source, pi0
主要產品
- pi0: Vision-Language-Action foundation model for robot control
- pi0.5: VLA with open-world generalization to new environments
- pi0-FAST: Autoregressive VLA variant using FAST tokenizer
- openpi: Open-source framework for training and deploying VLA models
成就
- First general-purpose robot foundation model: Created pi0, the first VLA model capable of controlling diverse robots across manipulation tasks
- Open-sourced VLA models: Released models and training code with 10k+ GitHub stars
- Massive pre-training dataset: Pre-trained on 10,000+ hours of robot data across multiple embodiments
- Zero-shot generalization: Demonstrated generalization to entirely new environments with pi0.5
募資
- 募資總額: $400M+
- 最近一輪: Series A
- 投資人: Bezos Expeditions, Thrive Capital, Lux Capital