A Humanoid Robot Understands Speech, Hands a Man an Apple, and Explains Why

In a two-minute video from the startup Figure, a robot running OpenAI's technology is asked for 'something to eat' and hands over the only food on the table, then explains its reasoning while clearing dishes. Nothing in the demo was scripted, the company says.

Figure, a California startup that has been building a humanoid robot for less than two years, posted a video today that has been watched several million times. A man stands at a counter with the robot, which is roughly his height. He asks what it sees; it describes the apple, the dish rack, the cups. He asks for something to eat, and it hands him the apple. He asks it, while it collects trash into a bin, why it did that. "I gave you the apple because it's the only edible item I could provide you with," it says, in a slightly halting voice, still sorting dishes.

The robot's understanding comes from OpenAI, which invested in the company two weeks ago and whose programs turn the camera images and the spoken questions into decisions about what to do. Figure's own software turns those decisions into movements. The company says the video is not sped up and not scripted — the robot was not told what would be asked.

There are reasons for caution. It is one video, chosen by the company. The tasks are simple. The robot is slow. Demonstrations of humanoid robots have a long history of looking better than the product. But what the video shows had not been shown before: a robot that can be talked to like a person, in a room it wasn't programmed for, and that can say why it did what it did.

Figure has raised $675 million from investors including Microsoft, Nvidia, and Amazon's founder, Jeff Bezos, and says it has an agreement to put robots in a BMW factory. Two years ago it did not exist.