Listwise Reward Estimation for Offline Preference-based Reinforcement Learning

TL;DR

This paper introduces LiRE, a novel offline preference-based reinforcement learning method that utilizes second-order preference information through ranked lists of trajectories, improving reward estimation accuracy and robustness.

Contribution

LiRE is the first approach to incorporate second-order preferences in offline PbRL, enhancing reward estimation by constructing ranked trajectory lists using ternary feedback.

Findings

LiRE outperforms state-of-the-art baselines in experiments.

LiRE is robust to feedback noise and varying feedback quantities.

A new offline PbRL dataset was proposed for evaluation.

Abstract

In Reinforcement Learning (RL), designing precise reward functions remains to be a challenge, particularly when aligning with human intent. Preference-based RL (PbRL) was introduced to address this problem by learning reward models from human feedback. However, existing PbRL methods have limitations as they often overlook the second-order preference that indicates the relative strength of preference. In this paper, we propose Listwise Reward Estimation (LiRE), a novel approach for offline PbRL that leverages second-order preference information by constructing a Ranked List of Trajectories (RLT), which can be efficiently built by using the same ternary feedback type as traditional methods. To validate the effectiveness of LiRE, we propose a new offline PbRL dataset that objectively reflects the effect of the estimated rewards. Our extensive experiments on the dataset demonstrate the superiority of LiRE, i.e., outperforming state-of-the-art baselines even with modest feedback budgets and enjoying robustness with respect to the number of feedbacks and feedback noise. Our code is available at https://github.com/chwoong/LiRE