Distributed Resource Allocation for Human-Autonomy Teaming With Human Preference Uncertainty

This paper investigates distributed resource allocation involving multiple autonomous agents and multiple humans, focusing on two challenges: (i) the dependency between autonomous and human agents through interaction; (ii) accounting for human uncertainties where both parties must collectively satisfy globally coupled probabilistic resource constraints. To address these, we first quantify the distribution of human choice behaviors using the maximum likelihood estimation (MLE), where human decisions evolve in response to nearby agent behaviors. Building on this human model, we introduce a novel reformulation that approximates the original probabilistic constraint via a polyhedral inner approximation, which then enables a fully distributed algorithm design over the systems communication graph while ensuring probabilistic constraint satisfaction. The proposed approach is validated through theoretical analysis and human-subject experiments.