Beating Atari with Natural Language Guided Reinforcement Learning

2017/04/18 by Russell Kaplan, Christopher Sauer, Kaplan, Russell +3 · 1 voice · 5 citations
Computer Science · #Reinforcement Learning in Robotics #Evolutionary Algorithms and Applications #Multimodal Machine Learning Applications

paper · pdf · doi:10.48550/arxiv.1704.05539

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

• HERAKLES: Hierarchical Skill Compilation for Open-ended LLM Agents
• Document-editing Assistants and Model-based Reinforcement Learning as a Path to Conversational AI
• Sample-Efficient Model-based Actor-Critic for an Interactive Dialogue\n Task
• DAIL: Beyond Task Ambiguity for Language-Conditioned Reinforcement Learning
• ACTRCE: Augmenting Experience via Teacher's Advice For Multi-Goal Reinforcement Learning

Discussions

• Beating Atari with Natural Language Guided Reinforcement Learning [pdf] [hn, 82 points, 12 comments]

Related

• A Review for Deep Reinforcement Learning in Atari:Benchmarks, Challenges, and Solutions
• Transferring Deep Reinforcement Learning with Adversarial Objective and Augmentation
• Is Deep Reinforcement Learning Really Superhuman on Atari? Leveling the playing field
• Playing Atari Games with Deep Reinforcement Learning and Human Checkpoint Replay
• Agent57: Outperforming the Atari Human Benchmark