Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Molins, Pau de las Heras, Yalcinkaya, Beyazit, Peters, Lasse, Fridovich-Keil, David, Bakirtzis, Georgios
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2605.10585
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866914553972916224
author Molins, Pau de las Heras
Yalcinkaya, Beyazit
Peters, Lasse
Fridovich-Keil, David
Bakirtzis, Georgios
author_facet Molins, Pau de las Heras
Yalcinkaya, Beyazit
Peters, Lasse
Fridovich-Keil, David
Bakirtzis, Georgios
contents Multi-objective reinforcement learning (MORL) allows a user to express preference over outcomes in terms of the relative importance of the objectives, but standard metrics cannot capture whether changes in preference reliably change the agent's behavior in the intended way, a property termed controllability. As a result, preference-conditioned agents can score well on standard MORL metrics while being insensitive to the preference input. If the ability to control agents cannot be reliably assessed, the symbolic interface that MORL provides between user intent and agent behavior is broken. Mainstream MORL metrics alone fail to measure the controllability of preference-conditioned agents, motivating a complementary metric specifically designed to that end. We hope the results spur discussion in the community on existing evaluation protocols to consolidate advances in preference adaptation in MORL to larger and more complex problems.
format Preprint
id arxiv_https___arxiv_org_abs_2605_10585
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Controllability in preference-conditioned multi-objective reinforcement learning
Molins, Pau de las Heras
Yalcinkaya, Beyazit
Peters, Lasse
Fridovich-Keil, David
Bakirtzis, Georgios
Machine Learning
Multi-objective reinforcement learning (MORL) allows a user to express preference over outcomes in terms of the relative importance of the objectives, but standard metrics cannot capture whether changes in preference reliably change the agent's behavior in the intended way, a property termed controllability. As a result, preference-conditioned agents can score well on standard MORL metrics while being insensitive to the preference input. If the ability to control agents cannot be reliably assessed, the symbolic interface that MORL provides between user intent and agent behavior is broken. Mainstream MORL metrics alone fail to measure the controllability of preference-conditioned agents, motivating a complementary metric specifically designed to that end. We hope the results spur discussion in the community on existing evaluation protocols to consolidate advances in preference adaptation in MORL to larger and more complex problems.
title Controllability in preference-conditioned multi-objective reinforcement learning
topic Machine Learning
url https://arxiv.org/abs/2605.10585