From 300 matchboxes learning tic-tac-toe to humanoid robots anticipating soccer shots, reinforcement learning reveals a ...