Virtual Leader-follower based Platooning Under Mixed Traffic: A Data-driven Distributionally Robust MPC Method

The trajectory prediction information of human-driven vehicles (HDVs) is imperative for the platooning of connected and automated vehicles (CAVs) in mixed traffic flow. However, due to the inherent uncertainty of HDVs, prediction errors are inevitable and time-varying, which results in inefficient control actions and even safety issues. To address the problem, we propose a virtual leader-follower platooning strategy to deal with time-varying prediction uncertainties. The controlled platoon tracks the efficient trajectory planned by the virtual leader to avoid unnecessary control inputs and improve speed smoothness. The challenges mainly lie in the modeling of time-varying prediction uncertainties and the design of control strategy for the virtual leader. First, a second-order moment based ambiguity set is employed to describe the distributional uncertainties of trajectory prediction, with parameters identified and updated by the Gaussian Mixture Model-Hidden Markov Model (GMM-HMM) to adapt to the time-varying environments. Further, a distributionally robust model predictive control (DR-MPC) method for the virtual leader is designed, where conditional value-at-risk constraint is adopted to achieve flexible tracking. By exploiting constraint tightening technology, the efficient solution of the DR-MPC problem is obtained. Simulation results demonstrate that our proposed method can significantly avoid undesired control inputs and improve driving smoothness.