Reinforcement is a decades-old way to let a computer learn to do something through experimentation combined with positive or negative feedback. It has emerged over the past decade when Google Deepmind showed that it could produce algorithms that could be to a superhuman strategy and game. More recently, AI engineers have used the technique to get great language models to act.

Raibert says many accurate new simulations have speeded up a difficult learning process by allowing robots to practice their movements in silico. ‘You don’t have to get so much physical behavior from the robot [to generate] Good performance, ”he says.

Several academic groups have published work that shows how reinforcement learning can be used to improve the movement of the bones. A team at UC Berkeley used the approach to train a humanoid to walk on their campus. Another group at Eth Zurich uses the method of leading four -legged to treacherous land.

Boston Dynamics has been building bones robots for decades, based on Raibert’s groundbreaking insights on how animals balance dynamically using the type of low-level control provided by their nervous system. However, as pleasant foot as the machine’s machines are more advanced behavior, including dance, parkour doing, and simply navigating in a room usually requires careful programming or some kind of human remote control.

In 2022, Rabert founded the robotics and AI (Rai) institute to investigate ways to increase the intelligence of bones and other robots so that they could do more on their own. While we wait for robots to actually learn how to do the dishes, they have to make fewer accidents. “You break fewer robots if you are going to manage the thing on the physical machine,” says Al Rizzi, chief technology officer at the Rai Institute.

What do you make of the many humanoid robots that are being demoted now? What kinds of tasks do you think they should do? Write to us at [email protected] or comments below.

Correction: 2/27/2025, 12:00 AM EDT: Marc Raibert’s title and certain biographical details were corrected, and wired the relationship between the businesses he founded and further clarified the progress in machine learning.