Physical Intelligence

Physical Intelligence (π) is a robotics AI company developing general-purpose foundation models intended to control any robot for any task. Its model series, including π0, π0.6 and π0.7, consists of vision-language-action (VLA) models trained on robot data, with research also exploring transfer from human video. The company publishes research on topics such as reinforcement learning from experience, long and short-term memory for robots, and precise manipulation. In October 2026 it opened a beta of the Pi API, which lets users upload data, fine-tune models and run cloud-hosted inference on their own robots.

• General-purpose vision-language-action robot models (π0, π0.6, π0.7)
• Models designed to work across different robot types and tasks
• Pi API beta: data upload, fine-tuning and cloud inference
• Published research on RL, robot memory and dexterous manipulation