Physical Intelligence is building pi-zero into a generalist robot policy
An October 31, 2024 pi-zero release puts Physical Intelligence around a robot policy trained across images, text, and actions.
Physical Intelligence published pi-zero on October 31, 2024 as its first generalist robot policy. The model spans images, text, and actions, then outputs low-level motor commands, placing the company in the robot foundation-model race rather than the robot hardware market.
The pi-zero release says the policy can control different robots and can be prompted or fine-tuned for harder task scenarios. The demonstrations include folding a shirt and cleaning a table, useful examples because they involve contact, deformable objects, and task sequencing.
Physical Intelligence was founded by a group of robotics and AI researchers and operators, including Sergey Levine and Lachy Groom. The company?s thesis is that generalist policies can become the control layer across robot bodies, but the proof has to move from demonstrations into reliable work in real customer environments.
The competitive field includes Skild AI, Sereact, Covariant RFM work, Google DeepMind robotics, NVIDIA GR00T, Figure?s Helix-style stack, Genesis AI, and internal foundation-model teams at robot OEMs. Physical Intelligence?s distinction is direct policy learning for low-level robot actions rather than only high-level task planning.
Public material does not show customer deployment count, task success rate by site, intervention frequency, robot-body transfer cost, field uptime, pricing, customer retention, paid deployment count, or renewal rate. The evidence is research artifact and demonstration breadth, not customer operations.
Pi-zero tests whether generalist robot policies can make robot capability portable across bodies and tasks. If Physical Intelligence converts model demonstrations into reliable customer task performance, it can position itself as the control layer that robot builders buy before deciding which body should do the work.
Have a robotics update Korthos should review? Send news, deployments, product releases, funding rounds, research, or media to tips@korthos.xyz or reach out on X at @agkorthos.