China Mechanical Engineering ›› 2026, Vol. 37 ›› Issue (8): 1900-1908.DOI: 10.3969/j.issn.1004-132X.2026.08.009

Previous Articles    

Research on Lateral Control of Intelligent Vehicles Based on Physics-Learning Hybrid Dynamics Modeling

SHI Peicheng1(), SUN Yuchen1, CHAKIR Chadia1, SUN Yu2   

  1. 1.School of Mechanical and Automotive Engineering,Anhui Polytechnic University,Wuhu,Anhui,241000
    2.Chery New Energy Vehicle Co. ,Ltd. ,Wuhu,Anhui,241002
  • Received:2025-03-26 Online:2026-08-25 Published:2026-09-17
  • Contact: SHI Peicheng

基于物理-学习混合动力学建模的智能车辆横向控制研究

时培成1(), 孙雨辰1, CHAKIR Chadia1, 孙羽2   

  1. 1.安徽工程大学机械与汽车工程学院, 芜湖, 241000
    2.奇瑞新能源汽车股份有限公司, 芜湖, 241002
  • 通讯作者: 时培成
  • 基金资助:
    中央引导地方科技专项-长三角科技创新共同体联合攻关计划(2023CSJGG1600);安徽省自然科学基金(2208085MF173);芜湖市“赤铸之光”重大科技项目(2023zd01);芜湖市“赤铸之光”重大科技项目(2023zd03)

Abstract:

To address the insufficient accuracy of traditional dynamic models for intelligent vehicle lateral control under complex operating conditions, this paper proposes a physics-based learning hybrid dynamic modeling method combined with learning-based MPC to enhance lateral control precision and robustness. A single-track model is first adopted as the foundational physical model and augmented with a gated recurrent unit (GRU) network to compensate for unmodeled dynamics, thereby constructing a hybrid model that integrates physical interpretability with data-driven adaptability. The hybrid model is then embedded as the predictive model within an MPC framework, in which an optimization objective function incorporating trajectory tracking errors and control input constraints is designed to compute optimal front -wheel steering angle commands in real time. CarSim/Simulink co-simulation results under double-lane- change and serpentine maneuvers demonstrate that the proposed method reduces both the root-mean-square lateral tracking error and the front-wheel steering angle amplitude compared with benchmark methods, while effectively suppressing unmodeled dynamics inherent in the single-track model. The proposed approach achieves precise control output under complex conditions, improving both lateral control accuracy and stability, and thus demonstrates superior lateral control performance for intelligent vehicles.

Key words: intelligent vehicle, hybrid model, model predictive control (MPC), lateral control

摘要:

针对智能车辆横向控制中传统动力学模型在复杂工况下精度不足的问题,提出了一种物理-学习混合动力学建模方法,并结合学习型模型预测控制方法以提高横向控制的精度与鲁棒性。首先以单轨模型为基础物理模型,引入门控循环单元(GRU)补偿未建模动态,构建混合物理学习模型,兼具物理可解释性与数据驱动自适应性;然后将混合模型作为预测模型嵌入模型预测控制(MPC)框架,设计考虑轨迹跟踪误差与控制量约束的优化目标函数,实时求解最优前轮转向角指令。CarSim/Simulink联合仿真结果表明,在双移线与蛇形工况下,相较于对比方法,所提方法有效降低了横向跟踪误差均方根,同时减小了前轮转角幅值,有效抑制了单轨模型中的部分未建模动态的影响。所提方法实现了车辆在复杂工况下控制量的精确输出,同步提升了智能车辆的横向控制精度和稳定性,具有良好的横向控制效果。

关键词: 智能车辆, 混合模型, 模型预测控制, 横向控制

CLC Number: