
Physical Intelligence
General-purpose AI models for robot control
CORE INFO
Physical Intelligence builds software that lets robots interpret camera input and natural-language instructions, then perform physical tasks such as folding laundry, packing boxes, and preparing food. Its API and hosted-inference products are aimed at robot makers and companies operating their own robots.
- Latest reported financing: at least $1 billion under discussion at an approximately $20 billion valuation, with Avenir and Lux Capital reportedly in talks to lead; the October 2026 round was not confirmed as closed.bloomberg.comqz.com
- Ultra deployment: π0.6 achieved 96.4% autonomy over a full packing shift at an Ultra customer site, according to Physical Intelligence.pi.website
- TechCrunch covered π0.7's ability to handle tasks it was not explicitly taught, while noting that the model still struggled with complex tasks from a single high-level instruction.techcrunch.com
WHY WE WOULD WORK AT PHYSICAL INTELLIGENCE
Build Physical Intelligence
Help pursue an ambitious mission: develop a model capable of controlling any robot across any task. Your work can move AI from screens into the physical world.
Advance Robot Foundation Models
Work on the π-series of vision-language-action models, combining visual input, natural-language instructions, and real-time motor control. The stack spans flow matching, imitation learning, reinforcement learning, and multimodal robotics.
Shape a New Platform
Contribute to openpi, the Pi API beta, and a broader intelligence layer for third-party robots. Your work could influence how robotics companies train, deploy, and improve their machines.
Learn Across Disciplines
Collaborate with engineers, scientists, roboticists, and company builders in a research-first environment. The team brings together expertise in robotics, reinforcement learning, hardware, software, and operations.
Create Real-World Impact
Help robots tackle practical tasks including laundry folding, grocery bagging, box assembly, food preparation, and warehouse packing. Public partner work with Weave Robotics and Ultra shows the technology moving beyond lab demos.
Join Strong Momentum
Physical Intelligence has raised $600 million in its Series B and grown to an estimated 200–336 employees. With π0.5, π*0.6, π0.7, openpi, and the Pi API beta, there is room to grow alongside an expanding technical and commercial effort.
MARKET AND TRACTION
MARKET POSITION
SERVICES
NOTABLE CUSTOMERS
KEY METRICS
✦ KEY METRICCOMPETITIVE ADVANTAGE
PRODUCT AND TECH
π-Series Robot Models
Physical Intelligence's π0, π0.5, π*0.6, and π0.7 are general-purpose vision-language-action models that interpret visual inputs and natural-language instructions to control different robots and perform physical tasks.
Open-Source Robotics Stack
openpi provides π0 code, model weights, fine-tuned checkpoints, inference examples, and support for platforms such as ALOHA and DROID. It enables researchers and developers to adapt the models to their own robots and tasks.
Pi API Beta
The waitlisted Pi API beta lets customers upload robot data, fine-tune Physical Intelligence models, and run cloud-hosted inference on customer-owned robots. It represents the company's emerging robotics AI infrastructure offering.
Robot Learning Technology
The company combines flow matching for continuous action generation with imitation learning, reinforcement learning, real-time action chunking, and cross-embodiment training. These methods support manipulation, generalization, and efficient adaptation across robot platforms.