Skip to content
embodied.guide

Back to the AI models

Physical IntelligenceUnited States2024

π0

An AI model: a vision-language-action model (VLA) takes camera images and an instruction and turns them into movements for a robot. This page describes the software, not a robot.

Robotsnot stated

Key figures as stated by the developer

6 of 6 stated
  • Model type ⁠B vision-language-action
  • Controls ⁠B arms and hands
  • Inputs ⁠A camera language robot state
  • Access ⁠C open download
  • Parameters ⁠A 3,300,000,000
  • Base model ⁠A PaliGemma
How to read the figures
14kgA30.9lbA
The letter in the grey square after a figure points to the source the figure came from. Robots with several sources carry several letters.
74.5lb
Imperial figures are our own conversion unless the manufacturer states an imperial figure itself. The source letter always belongs to the metric figure — a conversion has no source of its own.

The four states

Versions and notes in the key figures (6)

Model type

Developer wording: "we train our vision-language-action flow matching model".

Controls

The developer says the model is "learning to directly output low-level motor commands via a novel architecture". Paper: "These include single and dual-arm manipulators with 6-DoF and 7-DoF arms, as well as holonomic and nonholonomic mobile manipulators."

Inputs

Paper: "The observation consists of multiple RGB images, a language command, and the robot’s proprioceptive state"

Access

openpi README: "provide multiple base VLA model checkpoints". "By default, checkpoints are automatically downloaded from"

Parameters

"We add 300M parameters for the action expert (which is initialized from scratch) for a total of 3.3 billion parameters."

Base model

"The VLM backbone weights are initialized from PaliGemma".

Full specifications

9 of 12 fields stated
The letter points to the source at the foot of the page. Fields without a figure are dashed: the developer does not state them — that is not the same as zero, and not the same as no.
Field Value Source and caveats
Input and output
Model typevision-language-actionB

Developer wording: "we train our vision-language-action flow matching model".

Inputscamera language robot stateA

Paper: "The observation consists of multiple RGB images, a language command, and the robot’s proprioceptive state"

Controlsarms and handsB

The developer says the model is "learning to directly output low-level motor commands via a novel architecture". Paper: "These include single and dual-arm manipulators with 6-DoF and 7-DoF arms, as well as holonomic and nonholonomic mobile manipulators."

Degrees of freedom controllednot stated

Note The paper sizes its action vector for the largest robot in the training data: "to accommodate two 6-DoF arms, 2 grippers, a mobile base, and a vertically actuated torso". ⁠A
Size and speed
Parameters3,300,000,000A

"We add 300M parameters for the action expert (which is initialized from scratch) for a total of 3.3 billion parameters."

Control rate50HzB

Rate of motor commands, not of inference: "Dexterous robot manipulation requires π 0 to output motor commands at a high frequency, up to 50 times per second." Paper: "This enables our model to control robots at frequencies of up to 50 Hz for dexterous tasks such as laundry folding".
Where it runs and what it is built on
Runs wherenot stated

Note The openpi README supports remote inference: "the model can run on a different server and stream actions to the robot via a websocket connection". ⁠C

Base modelPaliGemmaA

"The VLM backbone weights are initialized from PaliGemma".
Openness
Accessopen downloadC

openpi README: "provide multiple base VLA model checkpoints". "By default, checkpoints are automatically downloaded from"

Licensenot stated

Note The openpi repository is marked "Apache-2.0 license" and also holds "LICENSE_GEMMA.txt", the Gemma Terms of Use; the README does not say which licence covers the π0 weights. ⁠D
The developer’s own claims
Learns new tasksneeds retrainingB

Developer claim: "More complex and dexterous tasks may require the model to be fine-tuned to specialize it to downstream challenges." "The model can then perform a wide variety of tasks, via either direct prompting or fine-tuning."

Works across robotsyesB

Developer claim: "It can control a variety of different robots".

Yes only if the developer itself says the model runs on different robot models, for example robots from two manufacturers or two of one manufacturer’s models. Several units of the same robot do not count.

Notes

  • Physical Intelligence describes π0 as a general-purpose robot foundation model, a vision-language-action flow matching model.
  • Physical Intelligence published the π0 blog post on 31 October 2024.
  • Country and city from the company registry CB Insights (396 Treat Avenue, San Francisco); the developer does not state its location.

Sources

One line per source, not per figure: a source has one retrieval date. The letter beside a figure points to the line it came from.