Design and experimental evaluation of an efficient MPC-based lateral motion controller considering path preview for autonomous vehicles. (June 2022)

• Record Type: Journal Article
• Authors: Chen, Guoying; Yao, Jun; Hu, Hongyu; Gao, Zhenhai; He, Lei; Zheng, Xiulei
• Is Part Of: Control engineering practice. Volume 123(2022)
• Publication Date: 2022-06
• Subjects: Autonomous vehicles -- Lateral motion control -- Model predictive control -- Path preview -- Computational efficiency; Automatic control -- Periodicals; 629.89
• DOI: 10.1016/j.conengprac.2022.105164
• Languages: English
• ISSNs: 0967-0661

Abstract

Abstract: Lateral motion control, a core autonomous driving technology, still faces the significant challenge of accurately tracking the reference path under complex and changeable driving maneuvers. In this regard, this study develops an efficient model predictive control (MPC)-based lateral motion controller considering path preview to improve the robustness and computational efficiency in large lateral acceleration or high-speed lateral motion control. As a typical model-based approach, the accuracy of the MPC predictive model substantially affects the controller's robustness. Thus, to improve the robustness of the lateral motion controller, a tire parameter online adaptive module (TPOAM) is proposed to update the MPC predictive model online to reduce the model mismatch. Unlike other online adaptation methods, the proposed TPOAM does not rely on complex and multi-parameter tire models or look-up tables. Through delay augmentation, the proposed method accurately models the steering system delay and compensates it in the MPC predictive model. In the engineering implementation of the MPC-based lateral motion controller, the long prediction horizon substantially deteriorates the computational efficiency of the receding optimization. Addressing this issue, the preview-follower theory is introduced into the predictive model to make full use of the future path information. The tracking deviation at the corresponding preview point can be obtained by previewing maneuvers similarly to human drivers at each time in the prediction horizon. This tracking deviation is also considered part of the MPC cost function. The total prediction horizon of the lateral motion controller considering path preview can be effectively extended while almost no computing time increment. The designed controller is verified on an autonomous vehicle platform in the low-speed large curvature and high-speed lane changing scenarios. Experimental results show that the proposed controller can effectively improve the robustness and computational efficiency of lateral motion control compared with the typically used MPC-based lateral motion approach.