MIT’s robotic cheetah taught itself how to run, set a new speed record in the process

MIT’s mini cheetah has taught itself how to run. The engineers behind this creation released a video showing off its new skills.

Engineers updated the way the cheetah ran, using using AI simulations instead of a human-designed controller. This has allowed it to better navigate tougher terrain and even break its speed record at 8.7 miles per hour, according to its project page. 

“This new learning-based approach lets the robot learn to run in a simulator, using a simple neural network as the controller,” the video explained.

While researchers are impressed that the robot learned to run on its own, it runs more like a newborn toddler on the playground than an actual cheetah in the wild. Overall, however, the AI simulations offered numerous benefits over human-designed controllers. One of those being it took just three hours to experience 100 days’ worth of virtual simulations.

The self-learning ability will allow the robot to quickly adopt any new surroundings. It taught itself to have baby steps on gravel, and also handled the sudden change in surface. A robot’s manual programming is time-consuming, but it learns everything up against failure, every time it encounters something new. 

ARTICLE: CHAITANYA DIVYESH PATEL

MANAGING EDITOR: CARSON CHOATE

PHOTO CREDITS: GIZMODO.COM