When to Trust Your Model: Model-Based Policy Optimization

TL;DR

This paper investigates when to trust models in reinforcement learning, proposing a simple, effective approach using short model-generated rollouts that improves sample efficiency and scales well.

Contribution

It introduces a practical method of using short model-generated rollouts in policy optimization, backed by theoretical analysis and empirical validation.

Findings

Surpasses prior model-based methods in sample efficiency

Matches the asymptotic performance of top model-free algorithms

Scales effectively to long horizons that challenge other methods

Abstract

Designing effective model-based reinforcement learning algorithms is difficult because the ease of data generation must be weighed against the bias of model-generated data. In this paper, we study the role of model usage in policy optimization both theoretically and empirically. We first formulate and analyze a model-based reinforcement learning algorithm with a guarantee of monotonic improvement at each step. In practice, this analysis is overly pessimistic and suggests that real off-policy data is always preferable to model-generated on-policy data, but we show that an empirical estimate of model generalization can be incorporated into such analysis to justify model usage. Motivated by this analysis, we then demonstrate that a simple procedure of using short model-generated rollouts branched from real data has the benefits of more complicated model-based algorithms without the usual pitfalls. In particular, this approach surpasses the sample efficiency of prior model-based methods, matches the asymptotic performance of the best model-free algorithms, and scales to horizons that cause other model-based methods to fail entirely.