The Journey is the Reward: Unsupervised Learning of Influential Trajectories