Skill-Aligned Fairness in Multi-Agent Learning for Collaboration in Healthcare

TL;DR

This work tackles fairness in heterogeneous MARL for healthcare by addressing both workload balance and skill-task alignment. It introduces MARLHospital, a healthcare-inspired environment that models energy-constrained, order-dependent tasks and team compositions, and FairSkillMARL, which defines a composite fairness objective $L_3 = \alpha L_1 + (1-\alpha)L_2$ with a fairness-shaped reward $r_t = R(\boldsymbol{s}_t, \boldsymbol{a}_t) - \lambda L_3$. Empirically, CTDE methods (VDN/QMIX) generally outperform DTDE baselines in MARLHospital, and incorporating skill alignment ($L_2$) into fairness improves alignment and reduces workload disparity relative to workload-only baselines, sometimes at the cost of strict equality. The results demonstrate the practical value of balancing workload with expertise in safety-critical MARL and provide a reusable benchmarking platform for studying fairness in heterogeneous teams. These insights pave the way for robust, healthcare-aware, multi-agent coordination that mitigates burnout and improves task performance.

Abstract

Fairness in multi-agent reinforcement learning (MARL) is often framed as a workload balance problem, overlooking agent expertise and the structured coordination required in real-world domains. In healthcare, equitable task allocation requires workload balance or expertise alignment to prevent burnout and overuse of highly skilled agents. Workload balance refers to distributing an approximately equal number of subtasks or equalised effort across healthcare workers, regardless of their expertise. We make two contributions to address this problem. First, we propose FairSkillMARL, a framework that defines fairness as the dual objective of workload balance and skill-task alignment. Second, we introduce MARLHospital, a customizable healthcare-inspired environment for modeling team compositions and energy-constrained scheduling impacts on fairness, as no existing simulators are well-suited for this problem. We conducted experiments to compare FairSkillMARL in conjunction with four standard MARL methods, and against two state-of-the-art fairness metrics. Our results suggest that fairness based solely on equal workload might lead to task-skill mismatches and highlight the need for more robust metrics that capture skill-task misalignment. Our work provides tools and a foundation for studying fairness in heterogeneous multi-agent systems where aligning effort with expertise is critical.

Figures (14)

• Figure 1: MARLHospital Environment. The environment integrates a PDDL planner with a MARL state layer to model skill-aligned fairness and shared-task coordination among healthcare workers. The goal is to pick the backboard from the crash cart, move to the patient, place it under the patient, compress patient chest multiple times, retrieve the Bag-Valve-Mask (BVM) from the crash cart, give patient rescue breaths multiple times.
• Figure 2: MARLHospital Execution flow.
• Figure 3: Task flow diagrams for the check compression (CPR) and rescue breath tasks in MARLHospital. The yellow-shaded subtask supports shared action with energy constraints.
• Figure 4: Test cumulative rates for the "chest compression" and "rescue breaths" tasks, with and without energy.
• Figure 5: $L_3$ composite disparity over training timesteps for different algorithms and skill configurations with no energy. Lower values indicate fairer agent behavior.