Beating Atari with Natural Language Guided Reinforcement Learning

TL;DR

This paper presents a novel deep reinforcement learning agent that leverages natural language instructions to improve performance in Atari games, especially Montezuma's Revenge, outperforming existing methods.

Contribution

It introduces a multimodal embedding approach enabling RL agents to use natural language guidance, enhancing learning efficiency and success in complex Atari environments.

Findings

Outperforms DQNs and A3C agents on Montezuma's Revenge

Uses natural language instructions to guide learning

Achieves higher scores than state-of-the-art agents

Abstract

We introduce the first deep reinforcement learning agent that learns to beat Atari games with the aid of natural language instructions. The agent uses a multimodal embedding between environment observations and natural language to self-monitor progress through a list of English instructions, granting itself reward for completing instructions in addition to increasing the game score. Our agent significantly outperforms Deep Q-Networks (DQNs), Asynchronous Advantage Actor-Critic (A3C) agents, and the best agents posted to OpenAI Gym on what is often considered the hardest Atari 2600 environment: Montezuma's Revenge.