Reinforcement learning

The robot tries actions, gets a score and slowly finds what works. It is the main method behind stable walking and running of legged robots, because nobody can hand-code every muscle correction needed to keep balance on uneven ground.

It needs a huge number of attempts, so it is done in simulation first (see sim-to-real). For you as a buyer, it explains why a modern robot dog recovers after a push or a slip, and why behavior can improve through software updates.