Physical Intelligence
AI robotics company building embodied general intelligence systems.
Founded
2024AI robotics company building embodied general intelligence systems.
Physical Intelligence pairs foundation-model methods with large-scale multi-task and multi-robot data collection plus a purpose-built architecture for dexterous manipulation, producing generalist policies that extend from pi0 to open-world pi0.5 and steerable pi0.7. Its research program compounds that lead across memory for long tasks, online reinforcement learning for precision, and efficient tokenization that trains generalist policies faster.
The company converts research into distribution on two tracks at once: open-sourcing pi0 weights, code, and the pi0-FAST model while placing partner deployments through a physical intelligence layer. That combination of top robot-learning founders, frontier generalist results, open ecosystem building, and deep venture backing is difficult for single-task automation vendors to replicate.
Source: pi.website
The market for generalist vision-language-action policies is expanding as manufacturers, logistics operators, and research labs look beyond single-task robots toward one model that controls many embodiments. Physical Intelligence positions its policy lineage from pi0 through pi0.5 and pi0.7 as that general layer, with open releases seeding developer adoption and partner deployments carrying models into production tasks such as cleanup, precise manipulation, and long-horizon work.
Competition is intensifying from closed frontier efforts such as Google DeepMind Gemini Robotics and from open baselines such as OpenVLA and Octo that buyers can finetune themselves. The outlook favors vendors that combine frontier generalization results with efficient training and deployment: open weights win mindshare while enterprise partners pay for reliability, safety, and throughput on real hardware.
Source: pi.website