There are some movies where we get comical scenes of people pushing over robots who are unable to pick themselves up after they fall. This seemed like a good way to run away from our robot overlords, at least until now where researchers have actually taught a robot dog how to pick itself back up after it has fallen over.
In the study published in the Science Robotics journal (via Engadget), researchers trained a neural network through a computer simulation. They then practiced it on a robot dog by literally kicking it over multiple times to test their training, and sure enough it worked because the robot dog eventually learned how to pick itself up.
According to the researchers, these moves were taught through the simulation which means that the ability to pick itself up is based completely on how the robot learned, not through specific maneuvers that were preprogrammed or built into the robot beforehand. The reason for the use of simulation is because it is a faster and cheaper way of training.
It also makes it better at adapting itself to various situations which will come in handy if ever deployed in the real world, where the robots could pick themselves up regardless of terrain, or at least in theory.