Robot AI model
π0-FAST
π0-FAST is an autoregressive VLA using frequency-domain action tokenization, with base and DROID fine-tuned checkpoints in openpi. Repository code uses Apache-2.0; Gemma Terms of Use apply to the inherited Gemma model components.
Records last verified Oct 13, 2026 · Send a correction
Also known as pi0-FAST
- Organization
- Physical Intelligence
- Type
- VLA
- Released / introduced
- Jan 16, 2025
- Public weights
- Public weights
- Licence / scope
- Apache-2.0; Gemma Terms of Use
- Parameters
- Not disclosed
- Base model
- PaliGemma-3B
- Humanoid association
- None reported
- Inputs
- vision, language
- Outputs
- tokenized action sequences
Release dates can mark an announcement or paper introduction. Public checkpoints, code licences and weight terms are distinct; follow the publisher’s links for the stated scope.
Documented embodiments
Source-associated hardware; this list does not establish plug-and-play compatibility or commercial robot capability.
- DROID Franka setup
The public distribution includes pi0-FAST-DROID.
Disclosed training data
- Physical Intelligence robot demonstration mixture
More robot AI models
Evidence
Sources
- [s1] FAST: Efficient Action Tokenization for Vision-Language-Action Models ↗
Physical Intelligence · paper · published Jan 16, 2025 · accessed Oct 13, 2026
- [s2] openpi: models, checkpoints and training examples ↗
Physical Intelligence · official · accessed Oct 13, 2026
- [s3] openpi Apache License 2.0 ↗
Physical Intelligence · official · accessed Oct 13, 2026
- [s4] Gemma Terms of Use distributed with openpi ↗
Physical Intelligence / Google · official · published Feb 21, 2024 · accessed Oct 13, 2026
- [s5] FAST paper: PaliGemma-based autoregressive VLA architecture ↗
Physical Intelligence · paper · published Jan 16, 2025 · accessed Oct 13, 2026