Making boats fly by scaling Reinforcement Learning with Software 2.0