Reinforcement learning in continuous state and action spaces