Skip to content
embodied.guide

Back to the AI models

Physical IntelligenceUnited States2025

π0.5

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

5 of 6 stated · 1 gap
  • Model type ⁠C vision-language-action
  • Controls ⁠A arms and hands
  • Inputs ⁠A camera language robot state
  • Access ⁠B open download
  • Parameters not stated
  • 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 (5)

Model type

Developer wording: "a VLA with Open-World Generalization".

Controls

The same command also covers the torso lift and the base of the mobile manipulator: "directly commands target poses for the arms, gripper, and torso lift, and the target base velocities at 50 Hz (with action chunking)".

Inputs

Paper: "the images from all of the cameras and the robot’s configuration (joint angles, gripper pose, torso lift pose, and base velocity)"; "is the overall task prompt"

Access

openpi README: "an upgraded version of π₀ with better open-world generalization". "By default, checkpoints are automatically downloaded from"

Base model

"The model weights are initialized from a standard VLM trained on data from the web"; "for the 2B VLM initialized from PaliGemma weights". Blog post: "building on our vision-language-action (VLA) model π 0".

Full specifications

7 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-actionC

Developer wording: "a VLA with Open-World Generalization".

Inputscamera language robot stateA

Paper: "the images from all of the cameras and the robot’s configuration (joint angles, gripper pose, torso lift pose, and base velocity)"; "is the overall task prompt"

Controlsarms and handsA

The same command also covers the torso lift and the base of the mobile manipulator: "directly commands target poses for the arms, gripper, and torso lift, and the target base velocities at 50 Hz (with action chunking)".

Degrees of freedom controllednot stated
Size and speed
Parametersnot stated
Control rate50HzA

Rate of the commands to the robot, not of inference: "directly commands target poses for the arms, gripper, and torso lift, and the target base velocities at 50 Hz (with action chunking)".
Where it runs and what it is built on
Runs wherenot stated
Base modelPaliGemmaA

"The model weights are initialized from a standard VLM trained on data from the web"; "for the 2B VLM initialized from PaliGemma weights". Blog post: "building on our vision-language-action (VLA) model π 0".
Openness
Accessopen downloadB

openpi README: "an upgraded version of π₀ with better open-world generalization". "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 π0.5 weights. ⁠D
The developer’s own claims
Learns new tasksnot stated

Note The developer states a different goal: "its goal is not to accomplish new skills or exhibit high dexterity, but to generalize to new settings". ⁠C

Works across robotsyesA

Developer claim: "We conducted all of our experiments using two types of mobile manipulators." The blog post says the model can "transfer physical behaviors from other 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 introduces π0.5 as a new model that generalizes to entirely new environments; it builds on the π0 model.
  • Physical Intelligence published the π0.5 blog post on 22 April 2025.
  • 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.