MIT’s long-jumping, back-flipping mini cheetah knows how to run now.
The engineers behind the robot released a video showing its new skills on Thursday. In it, the cheetah is shown bandying about on a variety of surfaces including grass, ice, and gravel.
It is said in the video that the cheetah has learned how to run on its own with the help of AI simulations. This is an update to the robot’s previous human-designed controller that has enabled it to better navigate tougher terrain and even break its own speed record at a whopping 8.7 mph.
“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.
The results seem to be quite humorous. The robot runs more like a newborn toddler on the playground than a fast cheetah in the wild. Still, it’s amazing that it learned to do it on its own.
The AI simulations offer various advantages over human-designed controllers. For one, it drastically cuts down training time. In fact, it takes just three hours for the robot cheetah to experience 100 days’ worth of virtual simulations.
This new learning method will allow the robot to adapt to any new surroundings quickly and efficiently. For example, if humans need to explore an unknown environment that’s potentially hostile to our weak fleshy bodies, we can simply send in the cheetah, and it can be ready to handle whatever comes.