Surprise-Based Intrinsic Motivation for Deep Reinforcement Learning

TL;DR

This paper introduces a surprise-based intrinsic motivation approach for deep reinforcement learning, enabling agents to explore more effectively in complex, sparse reward environments by modeling transition probabilities and using surprise as an exploration incentive.

Contribution

The authors propose a novel intrinsic motivation method that uses transition model surprise to guide exploration, improving performance in high-dimensional, sparse reward tasks.

Findings

Agents successfully explore complex environments with sparse rewards.

The method outperforms heuristic exploration strategies in various tasks.

Approximations of surprise facilitate scalable exploration in deep RL.

Abstract

Exploration in complex domains is a key challenge in reinforcement learning, especially for tasks with very sparse rewards. Recent successes in deep reinforcement learning have been achieved mostly using simple heuristic exploration strategies such as $\epsilon$-greedy action selection or Gaussian control noise, but there are many tasks where these methods are insufficient to make any learning progress. Here, we consider more complex heuristics: efficient and scalable exploration strategies that maximize a notion of an agent's surprise about its experiences via intrinsic motivation. We propose to learn a model of the MDP transition probabilities concurrently with the policy, and to form intrinsic rewards that approximate the KL-divergence of the true transition probabilities from the learned model. One of our approximations results in using surprisal as intrinsic motivation, while the other gives the $k$-step learning progress. We show that our incentives enable agents to succeed in a wide range of environments with high-dimensional state spaces and very sparse rewards, including continuous control tasks and games in the Atari RAM domain, outperforming several other heuristic exploration techniques.