Physical Intelligence Publishes Humanoid API, Data
Robotics firm Physical Intelligence has unveiled real-world performance data for its 'Weave' and 'Ultra' humanoid robots through an API-first approach. The move reflects a market trend toward providing customers with transparent, granular data on metrics like task completion and safety incidents. This allows end-users to better monitor, orchestrate, and retrain fleets of embodied agents in production environments.
- Physical Intelligence's API provides access to a "physical intelligence layer" built on general-purpose foundation models, with versions including π0.5 and π0.6. In a live deployment packaging e-commerce orders for its partner, Ultra, the π0.6 model achieved 96.4% autonomy over a full shift. - The API-first approach treats the Application Programming Interface as a primary product, defining a contract for how the robots should behave before the application code is written. This strategy is designed to abstract the complexity of robotic tasks—like grasping and collision avoidance—into simpler commands, similar to how software APIs abstract complex computations. - This move comes amid a surge in venture capital for humanoid robotics, with Physical Intelligence raising $400 million at a $2 billion valuation. The funding landscape is highly competitive, highlighted by Figure AI's $1 billion funding round in 2025, which included corporate investors like Intel Capital, Nvidia, and Salesforce Ventures. - The API enables a Robotics-as-a-Service (RaaS) model, where customers can access automation through a subscription, lowering high upfront capital investment. The global RaaS market is projected to grow significantly, with one forecast estimating it will increase by $2.49 billion between 2023 and 2028. - While software and AI are advancing, significant hardware challenges remain for humanoid robots, including battery energy density, thermal management, and the mass production of high-precision components like actuators and bearings. The current competitive focus in the industry has shifted from pure hardware capabilities to the development of the "brain," or the embodied intelligence model that allows for autonomous decision-making. - Many robotics technologies are considered dual-use, with applications in both commercial and government sectors. Startups in this space often leverage non-dilutive capital from defense agencies like DARPA and the Defense Innovation Unit (DIU) to fund research and development.
