Decoupling Representation Learning from Reinforcement Learning

Abstract

In an effort to overcome limitations of reward-driven feature learning in deep reinforcement learning (RL) from images, we propose decoupling representation learning from policy learning. To this end, we introduce a new unsupervised learning (UL) task, called Augmented Temporal Contrast (ATC), which trains a convolutional encoder to associate pairs of observations separated by a short time difference, under image augmentations and using a contrastive loss. In online RL experiments, we show that training the encoder exclusively using ATC matches or outperforms end-to-end RL in most environments. Additionally, we benchmark several leading UL algorithms by pre-training encoders on expert demonstrations and using them, with weights frozen, in RL agents; we find that agents using ATC-trained encoders outperform all others. We also train multi-task encoders on data from multiple environments and show generalization to different downstream RL tasks. Finally, we ablate components of ATC, and introduce a new data augmentation to enable replay of (compressed) latent images from pre-trained encoders when RL requires augmentation. Our experiments span visually diverse RL benchmarks in DeepMind Control, DeepMind Lab, and Atari, and our complete code is available at https://github.com/astooke/rlpyt/tree/master/rlpyt/ul.

Improved related works and fixed code hyperlink

References in corpus (7)

• Bootstrap your own latent: A new approach to self-supervised Learning
• DeepMind Control Suite
• Rainbow: Combining Improvements in Deep Reinforcement Learning
• CARLA: An Open Urban Driving Simulator
• Reinforcement Learning with Unsupervised Auxiliary Tasks
• Contrastive Learning of Structured World Models
• Bootstrap Latent-Predictive Representations for Multitask Reinforcement Learning

Cited by in corpus (33)

• Contrastive Representation Learning: A Framework and Review
• Mastering Visual Continuous Control: Improved Data-Augmented Reinforcement Learning
• Reinforcement Learning with Prototypical Representations
• Stabilizing Deep Q-Learning with ConvNets and Vision Transformers under Data Augmentation
• Pretraining Representations for Data-Efficient Reinforcement Learning
• Learning Vision-Guided Quadrupedal Locomotion End-to-End with Cross-Modal Transformers
• Contrastive Behavioral Similarity Embeddings for Generalization in Reinforcement Learning
• Representation Matters: Offline Pretraining for Sequential Decision Making
• Baby Intuitions Benchmark (BIB): Discerning the goals, preferences, and actions of others
• Generalization in Reinforcement Learning by Soft Data Augmentation
• SECANT: Self-Expert Cloning for Zero-Shot Generalization of Visual Policies
• Learning Visual Robotic Control Efficiently with Contrastive Pre-training and Data Augmentation
• Beyond Fine-Tuning: Transferring Behavior in Reinforcement Learning
• Behavior From the Void: Unsupervised Active Pre-Training
• Why Generalization in RL is Difficult: Epistemic POMDPs and Implicit Partial Observability
• Training Larger Networks for Deep Reinforcement Learning
• Which Mutual-Information Representation Learning Objectives are Sufficient for Control?
• PlayVirtual: Augmenting Cycle-Consistent Virtual Trajectories for Reinforcement Learning
• Cross-Trajectory Representation Learning for Zero-Shot Generalization in RL
• Return-based Scaling: Yet Another Normalisation Trick for Deep RL
• Learning Markov State Abstractions for Deep Reinforcement Learning
• Decoupling Value and Policy for Generalization in Reinforcement Learning
• Provably Efficient Representation Selection in Low-rank Markov Decision Processes: From Online to Offline RL
• Learning from learning machines: a new generation of AI technology to meet the needs of science
• On The Effect of Auxiliary Tasks on Representation Dynamics
• Reinforcement Learning with Latent Flow
• Making Curiosity Explicit in Vision-based RL
• Pretrained Encoders are All You Need
• Bayesian Robust Optimization for Imitation Learning
• Towards robust and domain agnostic reinforcement learning competitions
• Towards Robust Bisimulation Metric Learning
• Continual Learning in Deep Networks: an Analysis of the Last Layer
• Unsupervised Skill-Discovery and Skill-Learning in Minecraft