Follow your Nose: Using General Value Functions for Directed Exploration in Reinforcement Learning