Learn what matters: cross-domain imitation learning with task-relevant embeddings

Neural Information Processing Systems 

We jointly train the learner agent's policy and learn