Exploration by Random Network Distillation

Code

• github.com/openai/random-network-distillation
  OfficialIn papertf★ 0
• github.com/rle-foundation/rlexplore
  jax★ 458
• github.com/alirezakazemipour/ppo-rnd
  pytorch★ 55
• github.com/uoe-agents/derl
  pytorch★ 27
• github.com/microsoft/strategically_efficient_rl
  tf★ 21
• github.com/riveSunder/carle
  pytorch★ 9
• github.com/kngwyu/intrinsic-rewards
  pytorch★ 8
• github.com/balloch/rl-exploration-transfer
  pytorch★ 4
• github.com/michalnand/reinforcement_learning
  pytorch★ 2
• github.com/jakegrigsby/supersonic
  tf★ 0

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.

Tasks

Benchmark Results

Reproductions