We’ve developed Random Network Distillation (RND), a prediction-based method for encouraging reinforcement learning agents to explore their environments through curiosity, which for the first time[1] exceeds average human performance on Montezuma’s Revenge. RND achieves state-of-the-art performance, periodically finds all 24 rooms and solves the first level without using demonstrations or having a
![Reinforcement Learning with Prediction-Based Rewards](https://cdn-ak-scissors.b.st-hatena.com/image/square/2378b17d633317c25f54f63c3d1347b4b47ee44d/height=288;version=1;width=512/https%3A%2F%2Fopenai.com%2Fcontent%2Fimages%2F2018%2F10%2FScreen-Shot-2018-10-31-at-9.06.21-AM-1.png)