MIT's long-jumping, back-flipping mini cheetah is at it again, and this time it teaches itself how to run.
The engineers behind the robot showed off their new skills in a video. The cheetah is shown bandying about on a variety of terrain, including grass, ice, and gravel.
The video says that the cheetah learned how to run on its own. This is an update to the robot's previous human-designed controller that has allowed it to better navigate tougher terrain and even break its own speed record at a whopping 8.7 mph, according to its project page.
This new learning-based approach lets the robot learn to run in a simulation, using a simple neural network as the controller.
To be honest, the results are hilarious. The robot is more like a baby on the playground than a wild animal. It's pretty awesome that it was able to do it on its own.
You can check out the results for yourself.
The benefits of the simulations are numerous. It cuts down on training time. It takes just three hours for the robot to experience 100 days worth of virtual simulations.
It's a lot like how Neo learned combat skills in The Matrix, but instead of kung fu, the robot learns how to walk a little bit less awkwardly over gravel.
The researchers say that the new learning method will allow the robot to quickly adapt to any new surroundings. If we need to explore an environment that may be hostile to our weak bodies, we can simply send in the cheetah and be confident that it will be able to take on any terrain challenges.
MIT's cheetah is becoming more sophisticated. When Skynet comes alive, we will all be at its mercy.
MIT taught itself how to run and set a new speed record.
MIT has taught its robot to Leap.
Are you interested in supporting clean energy adoption? At UnderstandSolar.com, you can find out how much money you could save if you switched to solar power. Futurism.com may receive a small commission if you sign up through this link.
All Rights Reserved © 2024
