Deep Reinforcement Learning from Policy-Dependent Human Feedback

Abstract

To widen their accessibility and increase their utility, intelligent agents must be able to learn complex behaviors as specified by (non-expert) human users. Moreover, they will need to learn these behaviors within a reasonable amount of time while efficiently leveraging the sparse feedback a human trainer is capable of providing. Recent work has shown that human feedback can be characterized as a critique of an agent's current behavior rather than as an alternative reward signal to be maximized, culminating in the COnvergent Actor-Critic by Humans (COACH) algorithm for making direct policy updates based on human feedback. Our work builds on COACH, moving to a setting where the agent's policy is represented by a deep neural network. We employ a series of modifications on top of the original COACH algorithm that are critical for successfully learning behaviors from high-dimensional observations, while also satisfying the constraint of obtaining reduced sample complexity. We demonstrate the effectiveness of our Deep COACH algorithm in the rich 3D world of Minecraft with an agent that learns to complete tasks by mapping from raw pixels to actions using only real-time human feedback in 10-15 minutes of interaction.

References in corpus (6)

• Adam: A Method for Stochastic Optimization
• Continuous control with deep reinforcement learning
• Guided Cost Learning: Deep Inverse Optimal Control via Policy Optimization
• Deep reinforcement learning from human preferences
• Control of Memory, Active Perception, and Action in Minecraft
• A Deep Hierarchical Approach to Lifelong Learning in Minecraft

Cited by in corpus (15)

• Scalable agent alignment via reward modeling: a research direction
• Human Engagement Providing Evaluative and Informative Advice for Interactive Reinforcement Learning
• A Survey on Interactive Reinforcement Learning: Design Principles and Open Challenges
• Widening the Pipeline in Human-Guided Reinforcement Learning with Explanation and Context-Aware Data Augmentation
• FRESH: Interactive Reward Shaping in High-Dimensional State Spaces using Human Feedback
• PEBBLE: Feedback-Efficient Interactive Reinforcement Learning via Relabeling Experience and Unsupervised Pre-training
• Imitation Learning: Progress, Taxonomies and Challenges
• Retrospective Analysis of the 2019 MineRL Competition on Sample Efficient Reinforcement Learning
• Skill Preferences: Learning to Extract and Execute Robotic Skills from Human Feedback
• X2T: Training an X-to-Text Typing Interface with Online Learning from User Feedback
• Robot Learning via Human Adversarial Games
• Avoiding Tampering Incentives in Deep RL via Decoupled Approval
• Learning Online from Corrective Feedback: A Meta-Algorithm for Robotics
• Battlesnake Challenge: A Multi-agent Reinforcement Learning Playground with Human-in-the-loop
• Teachable Reinforcement Learning via Advice Distillation