Sub-optimal Experts mitigate Ambiguity in Inverse Reinforcement Learning