π0
Open-source generalist robot foundation model — weights and code (the openpi repo) for fine-tuning a single policy to control different robot embodiments, from two-arm platforms to mobile manipulators, across diverse manipulation tasks.
What it does
- Robot control
Features 8
- Base model code and weightsCode and weights for running the base pre-trained model in the openpi repository.
- Fine-tuned checkpointsCheckpoints fine-tuned for simple tasks on widely available platforms such as ALOHA and DROID.
- Inference example codeExample code to run inference on real-world and simulated robot platforms.
- Fine-tuning codeCode for fine-tuning the base model for your own tasks and platforms.
- PyTorch portA PyTorch port of openpi, prepared by Hugging Face, for those who prefer PyTorch over JAX.
- Flow-matching action outputsAugments pre-trained vision-language models with continuous action outputs via flow matching, a variant of diffusion models.
- Text instruction followingFollows various text instructions after training on broad and diverse data.
- Multi-robot controlControls a variety of different robots, either prompted to carry out a task or fine-tuned to specialize it.
How it compares
-
Skild Brain
The race to build general-purpose robotic intelligence — the foundation on which more specialized applications can be built, much like the LLM models that captivated the world three years ago — is heating up. Pittsburgh-based Skild AI, founded in 2023, just this month raised $1.4 billion at a $14 billion valuation and is taking a notably different approach.
News reporting · 19 Sep 2026
Sources
- Pricing
-
Official documentation · 28 Aug 2026
Physical Intelligence's own announcement of the open-source release.
- Sold within
-
Official documentation · 30 Sep 2026
The openpi repository is published under the "Apache-2.0 license" and states "For all models, we provide base model checkpoints, pre-trained on 10k+ hours of robot data, and examples for using them out of the box or fine-tuning them to your own datasets." Installation instructions and checkpoint paths such as "gs://openpi-assets/checkpoints/pi0_base" are given.
- Deployment
-
Official documentation · 13 Oct 2026
Page: 'To run the models in this repository, you will need an NVIDIA GPU with at least the following specifications'.
- Interfaces
-
Official documentation · 14 Oct 2026
Page: 'We provide a client utility with minimal dependencies that you can easily embed into any robot codebase. / First, install the
openpi-clientpackage in your robot environment:'. - Modalities
-
Official documentation · 12 Oct 2026
Page: 'a flow-based vision-language-action model (VLA) / stream actions to the robot via a websocket connection'.
- Intended for
-
openpi release lets 'researchers and practitioners...run creative new experiments adapting pi0 to their own platforms' and 'anyone to experiment with fine-tuning'
Also from Physical Intelligence
-
Pi API API service
Beta API for Physical Intelligence's robotic foundation models: upload robot data, launch fine-tuning on compute the company manages, and run the fine-tuned model on your own robot through a hosted inference gateway; onboarding is by waitlist, in batches.
Open Physical Intelligence in the directory
Something wrong here? Send a correction — quote this product id: pi-zero.