Beating Atari with Natural Language Guided Reinforcement Learning

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.

Cited by in corpus (14)

• An Introduction to Deep Reinforcement Learning
• Informed Machine Learning -- A Taxonomy and Survey of Integrating Knowledge into Learning Systems
• Teaching Machines to Describe Images via Natural Language Feedback
• ACTRCE: Augmenting Experience via Teacher's Advice For Multi-Goal Reinforcement Learning
• PixL2R: Guiding Reinforcement Learning Using Natural Language by Mapping Pixels to Rewards
• Learning Language-Conditioned Robot Behavior from Offline Data and Crowd-Sourced Annotation
• Cross-Domain Perceptual Reward Functions
• Expert-augmented actor-critic for ViZDoom and Montezumas Revenge
• Zero-shot Task Adaptation using Natural Language
• Document-editing Assistants and Model-based Reinforcement Learning as a Path to Conversational AI
• An Overview of Natural Language State Representation for Reinforcement Learning
• Sample-Efficient Model-based Actor-Critic for an Interactive Dialogue Task
• Continual and Multi-task Reinforcement Learning With Shared Episodic Memory
• Grounding Complex Navigational Instructions Using Scene Graphs