Go to AAMAS 2026 Conference homepageMORL4Water: A Modular Multi-Objective Reinforcement Learning Toolkit for Water Resource Management
Keywords: Multi-objective reinforcement learning, applications, water management, simulations, sustainability, benchmarks
TL;DR: MORL4Water is a toolkit for realistic water management environments, enabling benchmarking of MORL algorithms; case studies show current methods underperform baselines but yield insights via solution-set analysis.
Abstract: Many real-world decision problems involve multiple conflicting objectives. Multi-objective reinforcement learning (MORL) extends standard reinforcement learning to optimize over multiple objectives simultaneously, producing sets of policies that capture different trade-offs. Despite recent algorithmic advances, MORL research is still constrained by simplified benchmarks that limit insights into real-world applicability. We introduce MORL4Water, a modular toolkit for constructing MORL environments in water resource management. Built on MO-Gymnasium, MORL4Water enables researchers to create realistic water management scenarios directly from data and supports systematic evaluation of MORL methods in high-impact domains. We demonstrate its use through two case studies—the Nile and Susquehanna rivers—where we benchmark several MORL algorithms against EMODPS, a domain-specific baseline. In addition to standard performance evaluation, we conduct a post-hoc analysis of the solution sets produced by these algorithms, highlighting differences in exploration, scalability, and the diversity of trade-offs. Our results show that while most state-of-the-art MORL algorithms still underperform compared to EMODPS, particularly in higher-dimensional settings, solution-set analysis provides valuable insights and establishes a foundation for more robust and impactful applications of MORL.
Area: Innovative Applications (IA)
Generative A I: I acknowledge that I have read and will follow this policy.
Submission Number: 631
Loading