Robotics Firms Develop AI Systems for Physical Tasks

Robotics companies are developing AI systems that allow machines to learn physical tasks by observing and interacting with their environment. The approach differs from earlier programmed robots by enabling machines to adapt through repeated attempts rather than fixed instructions.

Robots is using the technology for tasks such as emptying washing machines and washing dishes. The company, based in Munich, recently opened an office in Palo Alto, California. Physical Intelligence has demonstrated robots performing tasks including making espresso drinks, folding laundry, and assembling boxes. The company is funded in part by OpenAI.

Bezos has invested in Project Prometheus, a robotics laboratory, and Physical Intelligence. Former Amazon vice president Ethan Evans said Bezos is interested in technologies that could benefit future generations. Venture capitalist Bob Nelsen described the challenge as figuring out how to reinvent the physical world.

Bettina Schön-Behanzin, a vice president at Agile Robots, said robots are not programmed as much as they adapt by seeing and doing. She estimated that practical applications could appear in three to five years rather than a decade. Sergey Levine, co-founder of Physical Intelligence, said the systems can combine skills learned in different areas to solve new problems, such as making vegetable soup after learning related tasks.