“I spilled the coke on the table.” The robot receives this command, finds the coke can, throws it in the trash, then brings the sponge and cleans the mess. It comes after making the most reasonable plans on its own, without basic programming.
According to the New York Times (NYT) and WIRED, an artificial intelligence startup that develops the brains of various robots, Physical Intelligence, has raised $400 million. This is much more than the $70 million investment that the entire startup founded this year in the U.S. has received.
Major investors include Amazon founder Jeff Bezos, ChatGPT producer OpenAI, and venture capital firm Thrive Capital. Why has so much investment been focused on the eight-month basic model of the robot.
Physical intelligence creates “basic software” that can operate on all robots, instead of the traditional approach of making software for specific machines and specific tasks. CEO Carol Hausmann, who worked on robotics at Google for many years, Professor Sergey Levin, who studied robotics at the University of California, Berkeley, and Professor Chelsea Finn at Stanford University participated.
“What we make is not a specific robot’s brain,” Hausman said. “It’s a generalist brain that can control all robots.”
In their recently released paper, they showed how robots can perform tasks such as folding laundry, cleaning tables, and folding complex boxes with thin corrugated cardboard. In the video released, the mobile robot takes laundry from a washing machine and folds it on the table. “This is a very difficult task for the robot,” the company said. “A flat T-shirt on the table can be folded by repeating a predetermined action, but moving the tangled laundry pile is not enough because it crumples and twists in many ways.” “Until now, there has been no proven robot system that performs tasks with such complexity,” he continued. Although the robot was a little slow, it even succeeded in folding the laundry, ranging from shorts to T-shirts, neatly stacking them up. The robot also organized various types of trash and plates on the table. It categorized what needed to be thrown away into the trash bin and the dishes into the dishwasher. “As a result of various learning, the robot can use various new strategies, such as shaking off the dishes and putting the trash in the trash bin,” the company explained.
“Universal robots are still in their infancy and there is a long way to go,” Physical Intelligence said, making it difficult for them to predict the pace of development. “Early results have shown a hopeful future for a highly general-purpose robot model that implements unprecedented dexterity and physical ability,” he said. “We expect greater progress in all directions next year.”
“After seeing the demo robot, I wondered if AI technologies such as ChatGPT are ready to replace numerous physical tasks,” Wired said. “It could revolutionize factories and warehouses and help the economy greatly, but it could also spark widespread fear that AI will replace people.”
EJ SONG
US ASIA JOURNAL
