NIPS 2020 Reinforcement Learning Paper: Imitation Learning
We pick some papers about imitation learning in nips 2020 to annotate.
Offline Imitation Learning with a Misspecified Simulator
This work investigate policy learning in the condition of a few expert demonstrations and a simulator with misspecified dynamics.
Story: There are two current approaches to apply RL in real-world without a costly trial-and-error process: imitation learning and to train a policy in a simulator. This paper images a feasible scenario that we have a few expert demonstrations and a simulator with misspecified dynamics.