Guiding Reinforcement Learning with Incomplete System Dynamics