Adaptive Optimal Decision in Multi-Agent Random Switching Systems

Abstract

Random switching models have been widely used in areas of communication, physics and aerospace, to capture the random movement patterns of mobile agents. In this letter, we study the optimal decision-making problem for multi-agent systems governed by random switching dynamics. In particular, we develop a novel online optimal control solution that integrates the reinforcement learning (RL) with an effective uncertainty sampling method, called multivariate probabilistic collocation method (MPCM), to adaptively find the optimal policies for agents of randomly switching mobility. We also develop a novel estimator that integrates the unscented Kalman filter (UKF) and MPCM to provide online estimation solutions for these agents. Efficiency and accuracy of the proposed solutions are analyzed. A concrete communication and antenna control co-design problem for a multi-UAV network is studied in the end to illustrate and validate the results.