Vehicle parameters.

Vehicle lateral stability control under hazardous operating conditions represents a pivotal challenge in intelligent driving active safety. To address the issue of maintaining vehicle stability during emergency braking on roads with low and non-uniform adhesion, this paper proposes an intelligent integrated longitudinal and lateral stability control algorithm based on the Proximal Policy Optimization (PPO) algorithm. Firstly, high-fidelity models of electromechanical braking (EMB) and steer-by-wire (SBW) systems are constructed in Amesim by leveraging their dynamic characteristics, while a full-vehicle dynamics model is developed in CarSim. The dynamic accuracy of the drive-by-wire system is verified through input-output tracking analysis. Next, vehicle stability is analyzed using vehicle dynamics models to optimize reinforcement learning control variables. This involves designing a continuous state space and action space that incorporate vehicle states and road surface parameters. A multi-objective reward function is formulated using stability indicators, including critical tire slip angle, critical sideslip angle, and critical yaw rate thresholds. Training is conducted via an Amesim-CarSim-Python co-simulation platform for emergency braking scenarios on split-μ roads, low-adhesion surfaces, and curved roads. Results show that, compared to Model Predictive Control (MPC) and Sliding Mode Control (SMC), the PPO algorithm reduces braking distance by 15–20% on low-adhesion roads, decreases lateral deviation by 25–30% on split-μ roads, and suppresses yaw rate oscillation by 28.8% on curved roads. Hardware-in-the-loop (HIL) validation confirms the algorithm’s robustness under extreme conditions, with lateral stability metrics maintained within safety thresholds.危险工况下的车辆横向稳定性控制是智能驾驶主动安全领域的核心挑战。针对低附着且不均匀附着路面紧急制动时的车辆稳定性维持问题，本文提出一种基于近端策略优化（Proximal Policy Optimization，PPO）算法的智能集成式纵向与横向稳定性控制算法。首先，借助机电制动（Electromechanical Braking，EMB）和线控转向（Steer-by-Wire，SBW）系统的动态特性，在Amesim中构建其高保真模型，并在CarSim中搭建整车动力学模型。通过输入输出跟踪分析验证了线控系统的动态精度。随后，借助车辆动力学模型分析车辆稳定性以优化强化学习控制变量：设计包含车辆状态与路面参数的连续状态空间与动作空间；采用关键轮胎侧偏角、关键侧偏角与关键横摆率阈值作为稳定性指标，构建多目标奖励函数。在Amesim-CarSim-Python联合仿真平台上，针对分离μ路面、低附着路面与弯道场景下的紧急制动工况开展训练。结果表明，相较于模型预测控制（Model Predictive Control，MPC）与滑模控制（Sliding Mode Control，SMC），PPO算法在低附着路面可缩短制动距离15%~20%，在分离μ路面可降低横向偏差25%~30%，在弯道场景可抑制横摆率振荡28.8%。硬件在环（Hardware-in-the-Loop，HIL）验证证实了该算法在极端工况下的鲁棒性，其横向稳定性指标可维持在安全阈值范围内。