Inverse Reinforcement Learning infers a reward function from expert demonstrations, aiming to encode the behavior and intentions of the expert. Current approaches usually do this with generative and uni-modal models, meaning that they encode a single behavior. In the common setting, where there are various solutions to a problem and the experts sho...

Versatile Inverse Reinforcement Learning via Cumulative Rewards