tools
π0.5
π0.5 is a vision-language-action model for controlling mobile manipulators in unfamiliar homes and other environments.
π0.5 combines language, visual, and robotic-action data to control robots at both high and low levels. It can interpret broad instructions such as cleaning a bedroom, break them into subtasks, and produce motor commands.
Physical Intelligence evaluated it on tasks including putting dishes away, making beds, cleaning floors, rearranging objects, and wiping spills in homes absent from its training data. The model is not presented as a standalone chatbot or hosted consumer service; robot hardware, integration, deployment, and access arrangements are not listed.
Features
- Controls mobile manipulators in homes not seen during training
- Accepts high-level and detailed natural-language commands
- Breaks long tasks into semantic subtasks
- Outputs continuous low-level robot joint actions
- Uses a flow-matching action expert for motor control
- Co-trains on web, multimodal, and cross-robot data
- Handles object rearrangement and cleaning tasks
- Supports high- and low-level control in one VLA model
Use cases
- Clean kitchens and bedrooms in unfamiliar homes
- Put dishes and other objects in their intended places
- Make a bed from a high-level instruction
- Move specified objects into drawers
- Wipe spills with a sponge
- Adapt robot behavior to new environments
Pros
Cons
Latest updates
- Frontier Models for Every Robot
Models are available through the Pi API beta. Upload data, fine-tune models, and deploy inference on your own robots.
- π 0.7 : a Steerable Model with Emergent Capabilities (π 0.7)
Exhibits a step-change in generalization.
- Precise Manipulation with Efficient Online RL
We extract an RL Token from VLA models to enable online RL and improve throughput on tasks with few hours of data.
- VLAs with Long and Short-Term Memory
Multi-Scale Embodied Memory (MEM) gives our models memory and enables tasks longer than ten minutes.
- The Physical Intelligence Layer
General-purpose physical intelligence models will enable robotics applications. Our partners are solving real-world problems.
Capabilities
- Understands images — “Unlike LLMs, it spans images, text, and actions and acquires physical intelligence by training on embodied experience from robots, learning to directly output low-level motor commands via a novel architecture.” source
- Acts for you — “After post-training, the robot can unload the dryer, bring the clothes over to the table, and fold the clothes into a stack. The video is uncut, from a single policy operating fully autonomously.” source
Pricing
- Prices checked
- 2026-10-26