Skip to Main Content

The Self-Taught Running Robot

April 14, 2022

Thumbnail Photo courtesy of MIT CSAIL.


With the rise of automation and smart devices in all walks of modern day life, the effort to teach our machines more advanced tasks has increased exponentially. Our bots have mastered everything from knowledge keeping to heavy lifting, but there are still certain abilities that lie out of their reach, which is why the news that MIT has created a “mini cheetah” that taught itself to run is such an extraordinary accomplishment.

Rachel Gordon of MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) wrote in an article that “[r]esearchers from MIT’s Improbable AI Lab… have been working on fast-paced strides for a robotic mini cheetah — and their model-free reinforcement learning system broke the record for the fastest run recorded.”

Running has always proved to be a much harder activity than walking, picking up objects, or other physical activities since it requires so much fast-paced change. Maintaining speed while adjusting to new terrain (sudden ledges, concrete vs. grass, et cetera) has proven to not be so easy for our movement-designed robots.

Scientists and engineers dedicated to creating sprinting machines have previously relied on “analytical designs,” where it would be up to programmers to account for all potential changes and implement them into software. However, MIT’s mini cheetah is completely self-taught.

“Programming how a robot should act in every possible situation is simply very hard,” Gordon quotes developers Gabriel Margolis and Ge Yang. “[B]ecause 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. Learning by trial and error removes the need for a human to specify precisely how the robot should behave in every situation.”

By using reactive learning rather than preventative measures, the mini cheetah has significantly cut down on the amount of time it takes to adapt, “accumulat[ing] 100 days’ worth of experience on diverse terrains in just three hours of actual time.” Now the robot can run and spin over gravel and ice, climb down slippery hills and over ledges, and sprint at a rate of 3.9 meters per second.

With these feats, the mini cheetah has broken through several AI barriers and presents a new hope for the future of movement-based robots. “The traditional paradigm in robotics is that humans tell the robot both what task to do and how to do it… 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… 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 world where machines can be made to be every bit as physically capable as the average human may soon be upon us, and with it will remove tasks from the mundane to the perilous, saving lives through the precise manipulation of tools and the ability to overcome unpredictable scenarios.

Capitol Tech offers many opportunities in engineering, where you can design and build self-teaching robots just like MIT’s mini cheetah. To learn more about these programs, visit captechu.edu and peruse the various courses and degrees offered. Many courses are available both on campus and online. For more information, contact admissions@captechu.edu.