Hybrid Modeling of Heterogeneous Human Teams for Collaborative Decision Processes

The increasing integration of artificial intelligence (AI) enabled systems with human operators underscores the need for seamless collaboration across various domains. Accurate modeling of human behavior can enable AI-enabled systems to anticipate human decisions and align themselves to support humans in complex tasks. Unlike existing methods focusing primarily on individual human behavioral modeling, this paper models human behavior within heterogeneous teams working toward a cooperative objective. In such teams, members often have varying skills, knowledge, and levels of awareness, which influence their decision-making processes. This paper models team behavior as a sub-optimal, hybrid form of multi-agent reinforcement learning. By leveraging centralized training with hybrid centralized/decentralized execution, the model captures a spectrum of team behaviors, from fully centralized to fully decentralized and in between. This paper quantitatively models each team member’s awareness and communication levels, enabling the inverse learning of these parameters from observed human team behavior data. Numerical experiments validate the robustness and accuracy of the framework across diverse scenarios and team compositions, underscoring its effectiveness in modeling complex human interactions.