(ORDO NEWS) — MIT’s mini cheetah robot has broken its personal speed record by reaching 14.04 km/h thanks to a new reinforcement learning system.
This robot can not only run, but also learn new skills just by standing still. All thanks to a new machine learning platform
The mini cheetah is not the fastest four-legged robot. In 2012, its larger brother reached a top speed of 45.5 km/h, but the mini cheetah, being developed by MIT and the National Science Foundation Institute for AI and Fundamental Interactions, is much more flexible and able to learn to move without even having to try and fail.
In the new video, the four-legged robot can be seen crashing into barriers and then learning to jump over them, rushing over obstacles, jumping on one leg and adapting to slippery, icy terrain as well as loose gravel hills.
This adaptability is achieved through a simple neural network that can evaluate new situations that could put its hardware under a lot of stress.
How the mini cheetah works
Typically, the movement of a robot is controlled by a system that uses data based on the analysis of the movement of mechanical limbs to create models that serve as guides. However, these models are often inefficient and inadequate because it is impossible to foresee all the circumstances that may arise in reality.
When the robot runs at its maximum speed, it runs at the limit of its hardware, which makes it very difficult to simulate, so it is difficult for the robot to quickly adapt to sudden changes in the environment.
To overcome this problem, instead of analytically designed robots like Spot Boston Dynamics that rely on motion physics analysis and manual hardware and software tuning, the MIT team developed a robot learning platform by doing it by doing it.
This robot learns by trial and error without human intervention. If the robot has sufficient experience in different terrains, it can automatically improve its behavior.
And that experience doesn’t even have to be in the real world. Using simulation, the mini cheetah can accumulate 100 days of experience in three hours by standing still, the team says.
Contact us: [email protected]