Physical Intelligence is a robotics AI company founded in 2024 by Karol Hausman, Sergey Levine, and co-founders including Chelsea Finn, Lachy Groom, and Brian Ichter. The company develops learning algorithms and foundation models intended to control any robot to perform any task, bringing general-purpose artificial intelligence into the physical world through vision-language-action models.
Its research output centers on a family of generalist policies, including pi-zero and successive iterations such as pi-0.5 and pi-0.7, which aim to deliver open-world generalization and steerable behavior across robotic embodiments. The company has open-sourced foundational models and published work on real-time action chunking, efficient online reinforcement learning, and embodied memory, reflecting a research-forward orientation.
Physical Intelligence works with a small number of partner companies across verticals such as logistics, grocery, and manufacturing to test whether its systems can perform reliable real-world automation. Rather than rushing to productize, the company emphasizes building a general physical intelligence layer first, a strategy reflected in its substantial funding and high valuation backed by leading venture and strategic investors.
The company competes in an intensely active robot-learning field that includes both startups and major research labs. Its differentiation rests on founder pedigree from leading robotics and reinforcement learning research and on a clear bet that general models will outperform narrow automation in the long run.
Physical Intelligence is best understood as a frontier research company building foundational robot intelligence rather than a shipping product. Its near-term relevance is highest for robotics partners and researchers, with broad commercial deployment still developing.