Learning representations for control with hierarchical forward models

摘要: Learning control from pixels is difficult for reinforcement learning (RL) agents because representation learning and policy learning are intertwined. Previous approaches remedy …