Robust Mean-Field Games With Partial Observations: A Complementary Strategy

This work addresses a class of robust mean-field games for large-population multiagent systems, where each agent has access only to partial observations and is subject to an unknown bounded disturbance input. Unlike the existing literature in which formulated robust mean-field games are minimax problems, this work formulates a nonworst-case game problem and proposes a complementary control strategy with a decoupled design of mean-field tracking and robustness. A neat state-space realization for an operator $Q$ concerning robustness is provided, incorporating parameters of an $n$th-order $H_{\infty }$ controller. A consensus process example is provided to illustrate the robustness and performance of the proposed complementary mean-field control strategy.