MIT’s Robot ‘Mini Cheetahs’ Now Use AI to Navigate Terrain

Engineers at MIT have created a robotic “mini cheetah” that can run up to 9 mph and uses AI in order to navigate its terrain.

Engineers at MIT’s Improbable AI Lab, part of the Computer Science and Artificial Intelligence Laboratory (or CSAIL), have announced a new version of the private university’s viral-on-YouTube “mini cheetahs.” The mini cheetahs, which first went viral in early 2019—running and backflipping across campus (in the video at bottom)—now use an AI program that allows them to better navigate their terrain; one that, in fact, allows the robots to accumulate 100 days’ worth of experience on diverse terrains in just three hours of actual time.

“Programming how a robot should act in every possible situation is simply very hard… because if a robot were to fail on a particular terrain, a human engineer would need to identify the cause of failure and manually adapt the robot controller, and this process can require substantial human time,” MIT Assistant Professor and Project Director Pulkit Agrawal says in an MIT press release. “Learning by trial and error removes the need for a human to specify precisely how the robot should behave in every situation. This would work if: (1) the robot can experience an extremely wide range of terrains; and (2) the robot can automatically improve its behavior with experience.”

To solve this problem the engineers turned to machine learning—i.e. algorithms that allow a system to improve upon a given task by “learning” how to get better.

“We developed an approach by which the robot’s behavior improves from simulated experience, and our approach critically also enables successful deployment of those learned behaviors in the real world,” Agrawal says. “The intuition behind why the robot’s running skills work well in the real world is: Of all the environments it sees in this simulator, some will teach the robot skills that are useful in the real world. When operating in the real world, our controller identifies and executes the relevant skills in real-time.”

The result is a small-to-medium-dog-sized quadrupedal robot that’s able to run up to approximately 8.7 miles per hour—a new top-speed record for the bots—and “adapt to unstable terrain like gravel.” The fact the mini cheetahs base their movements on a learned “intuition” developed from large numbers of trial-and-error simulations means they can even figure out how to walk and balance if they have an inoperable leg.

“A more practical way to build a robot with many diverse skills is to tell the robot what to do and let it figure out the how,” Agrawal says. “Our system is an example of this. In our lab, we’ve begun to apply this paradigm to other robotic systems, including hands that can pick up and manipulate many different objects.” (A brief overview of that project is presented in the video immediately above.)

Note: This project is being supported by the Defense Advanced Research Projects Agency’s Machine’s Common Sense Program. According to its website DARPA’s Common Sense Program “seeks to address the challenge of machine common sense by pursuing two broad strategies [that] envision machine common sense as a computational service, or as machine commonsense services.” DARPA says it wants to develop an AI that learns like a child and is able to read the web like an all-knowing research librarian.

Feature image: MITCSAIL

(Visited 5 times, 1 visits today)

Accessibility Toolbar