A group of researchers has developed a new type of reactive locomotion system for robots. Known as rapid motor adaptation, it allows a robotic device to traverse various types of terrain, learning from previous experiences.
The robot, built by the Chinese startup Unitree, has four legs, moves like a dog and cannot see where he is walking. Instead, it advances by adjusting to the Unique features of the surface it is traversing.
Researchers designed the software as a self-learning system, that is, they placed a simulated version of the robot from Unitree in a variety of simulated environments, with initial robot training greatly reducing learning times.
After being fit, the robot was cast on different surfaces in a wide variety of real-world environments.
In one of the situations, the robot chose a path where it followed a rocky beach. In another case, he descended a small waterfall, instantly reacting to the sudden downhill dive.
The team also made the robot walk on slippery plastic to test your skills on these types of surfaces.
Researchers indicate that the new training technique is entirely based on trial and error. This approach allows for much more subtle reactions than other learning systems, writes the TechExplore.
The experts involved in the work suggest that the technology can be useful in search and rescue operations where the terrain is notoriously unpredictable.