Issues in Using Function Approximation for Reinforcement Learning - The Robotics Institute Carnegie Mellon University