EcoFair-CH-MARL: Scalable Constrained Hierarchical Multi-Agent RL with Real-Time Emission Budgets and Fairness Guarantees

A new research paper introduces EcoFair-CH-MARL, a multi-agent reinforcement learning (MARL) framework designed to tackle the complex trilemma of efficiency, sustainability, and fairness in maritime logistics. Developed by researcher Saad Alqithami, the system unifies three core innovations: a primal-dual budget layer that mathematically guarantees cumulative emissions stay under a set cap despite unpredictable weather and demand; a fairness-aware reward transformer that enforces 'max-min' cost equity, ensuring no single vessel or company in a heterogeneous fleet bears a disproportionate burden; and a two-tier hierarchical architecture that separates high-level strategic routing from real-time vessel control, allowing the system to scale linearly with the number of agents.

The framework's performance was validated using a high-fidelity digital twin simulation of a maritime network with 16 ports and 50 vessels, driven by real automatic identification system (AIS) data. Results showed a 15% reduction in emissions, a 12% increase in throughput, and a 45% improvement in cost fairness compared to state-of-the-art hierarchical and constrained MARL baselines. Notably, it also outperformed fairness-specific MARL models (like SOTO and FEN) on equity metrics like the Gini coefficient. The research provides new theoretical guarantees, establishing O(√T) regret bounds for both constraint violations and fairness loss.

With its modular design compatible with various AI learners, EcoFair-CH-MARL represents a significant step toward deploying large-scale, autonomous multi-agent systems in safety-critical, real-world domains like shipping and energy grids. It directly addresses the pressing need for AI solutions that are not only efficient but also provably compliant with tightening environmental regulations and socially responsible in their operational outcomes.