Boston Dynamics Led a Robot Revolution. Now Its Machines Are Teaching Themselves New Tricks

Learning reinforcement is a decade old way to have a computer Learn to do something through the test Combined with positive or negative reactions. It came out in the last decade Google showed Dipmind It can produce superhuman techniques and gameplay algorithms. Recently, AI engineers have used the strategy to get large language models to treat them.

Ribart says that very accurate new simulations have enhanced the robot a strict learning process by allowing them to practice their steps in silico. “You don’t have to get so much physical behavior from the robot [to generate] Good performance, “he said.

Several academic groups have revealed the job that shows how reinforcement can be used to improve legged locomotion. A team in UC Berkeley used this method Give them a humanoid training to visit the campusThe ETH is using another group of Zurich The quadrilaterals across the treacherous ground guideThe

Boston Dynamics has been creating leged robots for decades, based on Ribart’s pioneering insights, how animals make the lower-level control of the lower-level control provided by their nervous system. The company’s machines are at the foot, however, more advanced behaviors, including dancing, parko and simply navigating around a room, are usually required to carefully programming or any kind of human remote control.

Ribart founded in 2022 Robotics and AI (RAI) Institute Explore the way Increase the intelligence of the legged and other robots So that they can do more on themselves. Although we wait to learn how to do food robots, AI should make them prone to less accidents. “You break the low robot when you come to run the thing on the physical machine,” said Al Rizi, chief technology officer at the RAI Institute.

What do you do to demoid many humanoid robots now? What kind of work do you think you think? Write us at Hello @wired.com Or comments below.

Correction: 2/27/2025, 12:00 AM EDT: Mark Ribart’s title and some biographical details have been modified, and the relationship between its established companies and the progress in machine learning is further made clear.