Exploration by Random Network Distillation

Burda et al. 2018

Abstract

We introduce an exploration bonus for deep reinforcement learning methods that is easy to implement and adds minimal overhead to the computation performed. The bonus is the error of a neural network predicting features of the observations given by a fixed randomly initialized neural network. We also introduce a method to flexibly combine intrinsic and extrinsic rewards. We find that the random network distillation (RND) bonus combined with this increased flexibility enables significant progress on several hard exploration Atari games. In particular we establish state of the art performance on Montezuma’s Revenge, a game famously difficult for deep reinforcement learning methods. To the best of our knowledge, this is the first method that achieves better than average human performance on this game without using demonstrations or having access to the underlying state of the game, and occasionally completes the first level.

Motivation

• Success in RL requires a large number of samples of experience, so we need exploration methods that scale well.

Algorithm

• Introduces an exploration bonus that is simple to implement, works with high-dim observations, can be used with any policy optimization algorithm, and is efficient to compute (one forward pass of a neural network on a batch of experience).

• RND Exploration bonus:

  • A predictor network predicts the (deterministic) output of a fixed, randomly initialized target network

Discussion