Evolving Policies to Solve the Rubik's Cube: Experiments with Ideal and Approximate Performance Functions