reward learning

We propose a method for quantifying the similarity of learned reward functions without performing policy learning and evaluation.