Training Robots to Evaluate Robots: Example-Based Interactive Reward Functions for Policy Learning