π 0-FAST
π0-FAST (pi zero FAST) is an autoregressive Vision-Language-Action (VLA) robot foundation model developed by Physical Intelligence (π). It is a variant of the π0 generalist robot policy that uses the FAST tokenizer for autoregressive discretization of robot actions, enabling training of generalist policies approximately 5x faster than previous approaches. π0-FAST provides somewhat better language-following performance than the standard π0 flow-matching model, but with approximately 4–5x higher inference cost. Released as open-source weights via the openpi GitHub repository, it can be fine-tuned to custom robot platforms with 1–20 hours of demonstration data. Physical Intelligence, founded in San Francisco, has raised over $1B in funding at a $5.6B valuation.
Primary use case: General-purpose robot control policy for dexterous manipulation tasks; fine-tunable foundation model for researchers and developers building robot AI systems
Specifications
Physical
Locomotion
unknown
Manipulation
Hand Type
unknown
Power & Autonomy
Autonomy Level
semi_autonomous
AI & Compute
AI Capabilities
vision, planning, manipulation, learning
Commercial
Availability
research
Primary Use Case
General-purpose robot control policy for dexterous manipulation tasks; fine-tunable foundation model for researchers and developers building robot AI systems
Related news
Swiss startup Embodied AI launches AIR, a humanoid robot that learns from factory workers
XPeng open-sourced GroundingPI , a robot perception model that beats GPT-6 Astra
Figure's baby AGI, Meta's China gamble, China 5Xs physical AI investment, and Agility goes to Wall Street
Saturn Dynamics says its world model beats Nvidia's Cosmos by 73% ... and runs 38X faster
Washington targets robot imports, Agility goes public, $20 billion valuation, and home insurance for humanoids
Visual Point‐Line and Inertial Measurement Unit Fusion‐Based Simultaneous Localization and Mapping Algorithm for Robotic Grinding of Wind Turbine Blades
ABSTRACT Various drawbacks exist in traditional visual SLAM methods for wind turbine blade polishing scenarios. Insufficient feature extraction, unstable matching, and reduced positioning accuracy are easily produced. These issues are triggered by weak textures, strong reflections, and illumination fluctuations. A point-line fused visual-inertial positioning method named EPL-VINS is therefore proposed. FAST feature points and EDLines feature lines are adopted in the visual front end of this method. An adaptive threshold feature detection strategy and a texture complexity-based fea
Robot Olympics, Human Problems: The Humanoid That Ran Too Fast and Snapped in Half
Robot Olympics, Human Problems: The Humanoid That Ran Too Fast and Snapped in Half
Xiaomi's latest factory video hints at bigger ambitions for its humanoid robots
Xiaomi's latest factory video hints at bigger ambitions for its humanoid robots