pi*0.6 (pi-star 0.6) is a vision-language-action generalist robot foundation model by Physical Intelligence that learns from real-world experience via reinforcement learning. Using a method called RECAP (RL with Experience and Corrections via Advantage-conditioned Policies), it improves policy performance on tasks such as laundry folding and espresso making beyond the level achievable through demonstration data alone. It is built on a 5-billion-parameter vision-language backbone with an integrated action expert.
Primary use case: Reinforcement learning-trained generalist robot policy for dexterous manipulation tasks
Locomotion
unknown
Hand Type
unknown
Autonomy Level
semi_autonomous
AI Capabilities
vision, planning, manipulation, learning
Availability
research
Primary Use Case
Reinforcement learning-trained generalist robot policy for dexterous manipulation tasks
Robot AI Has No Internet to Learn From: Tutor Intelligence Raises $75M to Build One
Foundry Robotics has announced its oversubscribed $100 million Series A funding round led by Marlinspike Partners. Participating in the round are early backers Khosla Ventures and Red Glass Ventures, and new investors SE Ventures, Wing VC, and Axiom Ventures, as well as existing investors Hanabi and Zero Shot Fund. Foundry builds and operates autonomous factories […]
Jeff Bezos-backed Physical Intelligence is reportedly eyeing a $1 billion funding round. The available text provides no details about the round type, valuation, investors, or any associated robot models and specifications.