When you use reinforcement learning to teach a robot arm to "grasp an object" in a simulator, unbelievable things happen one after another. The more you continue learning, the more the success rate ...