Leveraging Human Guidance for Deep Reinforcement Learning Tasks

摘要: Reinforcement learning agents can learn to solve sequential decision tasks by interacting with the environment. Human knowledge of how these be incorporated using imitation learning, where agent learns imitate human demonstrated decisions. However, guidance is not limited demonstrations. Other types could more suitable for certain and require less effort. This survey provides a high-level overview five recent frameworks that primarily rely on other than conventional, step-by-step action We review motivation, assumption, implementation each framework. then discuss possible future research directions.