Revisiting Design Choices in Offline Model-Based Reinforcement Learning

TL;DR

This paper critically examines various heuristics for uncertainty in offline model-based reinforcement learning, revealing how hyperparameter optimization can significantly enhance performance over existing methods.

Contribution

It compares different uncertainty heuristics, introduces new protocols for their evaluation, and demonstrates the effectiveness of Bayesian Optimization in hyperparameter tuning for offline MBRL.

Findings

Bayesian Optimization outperforms hand-tuned hyperparameters.

Hyperparameters like model count and rollout horizon critically affect performance.

Selected hyperparameters lead to significantly stronger results.

Abstract

Offline reinforcement learning enables agents to leverage large pre-collected datasets of environment transitions to learn control policies, circumventing the need for potentially expensive or unsafe online data collection. Significant progress has been made recently in offline model-based reinforcement learning, approaches which leverage a learned dynamics model. This typically involves constructing a probabilistic model, and using the model uncertainty to penalize rewards where there is insufficient data, solving for a pessimistic MDP that lower bounds the true MDP. Existing methods, however, exhibit a breakdown between theory and practice, whereby pessimistic return ought to be bounded by the total variation distance of the model from the true dynamics, but is instead implemented through a penalty based on estimated model uncertainty. This has spawned a variety of uncertainty heuristics, with little to no comparison between differing approaches. In this paper, we compare these heuristics, and design novel protocols to investigate their interaction with other hyperparameters, such as the number of models, or imaginary rollout horizon. Using these insights, we show that selecting these key hyperparameters using Bayesian Optimization produces superior configurations that are vastly different to those currently used in existing hand-tuned state-of-the-art methods, and result in drastically stronger performance.