Improving Reinforcement Learning For Robotics With Control And Dynamical Systems Theory