In this interview, Sergey Levine discusses Physical Intelligence, a company aiming to build robotic foundation models capable of controlling any robot for any task. He shares the progress made in the first year, including robots performing dexterous tasks like folding laundry and cleaning kitchens. Levine outlines the year-by-year vision, emphasizing the challenges of enabling robots to understand and execute complex, long-duration tasks with common sense, continuous learning, and safety. He estimates that robots capable of doing useful tasks will be deployed in the real world within a few years, leading to a flywheel effect of continuous improvement through real-world experience. The conversation explores the differences between robotics and LLMs, the importance of human-in-the-loop systems, and the potential for robots to revolutionize industries by augmenting human productivity. Levine also touches on the challenges of scaling data collection, the role of simulation, and the hardware bottlenecks in robotics.
Sign in to continue reading, translating and more.
Continue