Meet Cassie, a bipedal robot that just completed a 5K race in 53 minutes. Slightly resembling a mini AT-ST Walker from the Star Wars movies, this computerized set of legs made history as the first untethered machine to do so on a single charge.
Not only that, Cassie taught itself to run. Using a “deep reinforcement” learning algorithm, the computer figured out how to stay upright by transferring weight from one leg to the other while on the move, Brian Heater reports for Tech Crunch.
“Deep reinforcement learning is a powerful method in AI that opens up skills like running, skipping and walking up and down stairs,” Yesh Godse says in a statement. Godse, an undergrad student at Oregon State University (OSU), was part of the team that put Cassie through its paces during the 5-kilometer trial.
The robot was invented by the university’s Dynamic Robotics Laboratory and produced by OSU spinoff company Agility Robotics. Working with a $1 million grant from the Defense Advanced Research Projects Agency of the United States Department of Defense, Oregon State robotics professor Jonathan Hurst led students in preparing Cassie for the historic run.
“The Dynamic Robotics Laboratory students in the OSU College of Engineering combined expertise from biomechanics and existing robot control approaches with new machine learning tools,” says Hurst, who is also a co-founder of Agility Robotics, in the press release. “This type of holistic approach will enable animal-like levels of performance. It’s incredibly exciting.”
The robot was able to remain standing for most of the run, which is a major achievement, Futurism’s Dan Robitzski reports. Many developers have had a difficult time keeping their robotic creations in an upright manner.
“Cassie is a very efficient robot because of how it has been designed and built, and we were really able to reach the limits of the hardware and show what it can do,” says Oregon State Ph.D. student Jeremy Dao, who works in the Dynamic Robotics Laboratory.
Cassie covered the course—slightly more than three miles—with a finishing time of 53:03. It would have completed the run faster, but a few glitches that added 6.5 minutes, reports James Vincent of The Verge. The bot fell twice during the experiment: once when the computer overheated and another time when its student handler directed it to take a turn too sharply