An Offline Adaptation Framework for Constrained Multi-Objective Reinforcement Learning

TL;DR

This work proposes a simple yet effective offline adaptation framework for multi-objective RL problems without assuming handcrafted target preferences, but only given several demonstrations to implicitly indicate the preferences of expected policies.

Abstract

In recent years, significant progress has been made in multi-objective reinforcement learning (RL) research, which aims to balance multiple objectives by incorporating preferences for each objective. In most existing studies, specific preferences must be provided during deployment to indicate the desired policies explicitly. However, designing these preferences depends heavily on human prior knowledge, which is typically obtained through extensive observation of high-performing demonstrations with expected behaviors. In this work, we propose a simple yet effective offline adaptation framework for multi-objective RL problems without assuming handcrafted target preferences, but only given several demonstrations to implicitly indicate the preferences of expected policies. Additionally, we demonstrate that our framework can naturally be extended to meet constraints on safety-critical objectives by utilizing safe demonstrations, even when the safety thresholds are unknown. Empirical results on offline multi-objective and safe tasks demonstrate the capability of our framework to infer policies that align with real preferences while meeting the constraints implied by the provided demonstrations.

Figures (14)

• Figure 1: Results on D4MORL Amateur datasets. Higher average utility and Hypervolume are preferable. The dashed lines represent the best performance between the original MODF and MORvS.
• Figure 2: Pareto fronts of different algorithms on D4MORL Amateur datasets. Each point represents an adapted policy for a specific unknown target preference.
• Figure 3: The comparison between the real target preferences and the adapted preferences.
• Figure 4: The adapted policies' cost and utility of each algorithm under various safety thresholds. Here, the utility is the normalized reward, since there is only one unconstrained objective in DSRL tasks. The points above the black dashed line represent the policies that violate the constraints.
• Figure 5: The maximum cost, the average utility and Hypervolume over all targets with a specific safety threshold.