Using a technique called reinforcement learning, a researcher at Google Brain has shown that virtual robots can redesign their body parts to help them navigate challenging obstacle courses—even if the solutions they come up with are completely bizarre.
