Hierarchical Deep Reinforcement Learning vs. Montezuma's Revenge

The authors were able to do well on one of the few games previously too difficult for Deep Reinforcement learning. The agent finds subgoals based on object detection.