Reinforcement Learning from Imperfect Demonstrations

Yang Gao, Huazhe Xu, Ji Lin +3

Robust real-world learning should benefit from both demonstrations and interactions with the environment. Current approaches to learning from demonstration and reward perform super…