Reinforcement Learning Using Local Adaptive Models