Using Imagery to Simplify Perceptual Abstraction in Reinforcement Learning Agents