Facebook has helped Unitree's robot dog technology learn to adapt to movement in any environment. The developers of the US company assure that Rapid Motor Adaptation (RMA) is the first such technology based on self-learning. Now robots explore the environment and interact with it, adapting to it and making independent decisions: how to proceed.
Previously, robotic dogs acted according to an algorithm: thanks to it, they could dance, overcome obstacles, climb stairs or bring any objects. However, now Belka and Strelka will have technologies with end-to-end training, during which robotic dogs will begin to independently understand exactly how to change the position of the limbs to continue moving - they have no prior knowledge. It will take a matter of fractions of a second to decide.
I immediately tested the technology in a real experiment, during which robotic dogs easily moved along sandy, rocky, or country roads, overcame areas with tall grass and other obstacles, for example, spilled oil. But even with this, the RMA algorithm coped with it.
Vivian Blumenthal expects the emergence of such technologies to lower the cost of robots by increasing their versatility. The ability to adapt to an unknown environment in real-time will be useful when performing various tasks, including when conducting search and rescue operations in areas potentially dangerous to people.