1. Why is the proposed approach able to learn a goal classifier from just a few examples of successful end states? (Just the main idea why, 1-2 sentences)
2. How is the resulting goal classifier used for planning or reinforcement learning? (2-3 sentences)