Investigating Pre-Training Objectives for Generalization in Vision-Based Reinforcement Learning
This work addresses the problem of evaluating generalization in vision-based RL for researchers, providing a benchmark and insights into pre-training objectives, though it is incremental in nature.
The paper tackles the unclear generalization ability of pre-training methods in vision-based reinforcement learning by introducing the Atari Pre-training Benchmark, which shows that task-agnostic objectives enhance generalization across diverse environments, while task-specific ones improve only in similar environments.
Recently, various pre-training methods have been introduced in vision-based Reinforcement Learning (RL). However, their generalization ability remains unclear due to evaluations being limited to in-distribution environments and non-unified experimental setups. To address this, we introduce the Atari Pre-training Benchmark (Atari-PB), which pre-trains a ResNet-50 model on 10 million transitions from 50 Atari games and evaluates it across diverse environment distributions. Our experiments show that pre-training objectives focused on learning task-agnostic features (e.g., identifying objects and understanding temporal dynamics) enhance generalization across different environments. In contrast, objectives focused on learning task-specific knowledge (e.g., identifying agents and fitting reward functions) improve performance in environments similar to the pre-training dataset but not in varied ones. We publicize our codes, datasets, and model checkpoints at https://github.com/dojeon-ai/Atari-PB.