The Surprising Ineffectiveness of Pre-Trained Visual Representations for Model-Based Reinforcement Learning

TL;DR

This paper evaluates the effectiveness of pre-trained visual representations in model-based reinforcement learning, finding they do not improve sample efficiency or out-of-distribution generalization as previously assumed.

Contribution

It provides a comprehensive benchmark of PVRs in MBRL, revealing their limited benefits and highlighting the importance of data diversity and architecture for generalization.

Findings

PVRs do not enhance sample efficiency in MBRL.

PVRs do not improve out-of-distribution generalization.

Data diversity and architecture are key for OOD performance.

Abstract

Visual Reinforcement Learning (RL) methods often require extensive amounts of data. As opposed to model-free RL, model-based RL (MBRL) offers a potential solution with efficient data utilization through planning. Additionally, RL lacks generalization capabilities for real-world tasks. Prior work has shown that incorporating pre-trained visual representations (PVRs) enhances sample efficiency and generalization. While PVRs have been extensively studied in the context of model-free RL, their potential in MBRL remains largely unexplored. In this paper, we benchmark a set of PVRs on challenging control tasks in a model-based RL setting. We investigate the data efficiency, generalization capabilities, and the impact of different properties of PVRs on the performance of model-based agents. Our results, perhaps surprisingly, reveal that for MBRL current PVRs are not more sample efficient than learning representations from scratch, and that they do not generalize better to out-of-distribution (OOD) settings. To explain this, we analyze the quality of the trained dynamics model. Furthermore, we show that data diversity and network architecture are the most important contributors to OOD generalization performance.