【作者】 姚军；

【导师】 陈国迎；

【作者基本信息】 吉林大学 ， 车辆工程， 2024， 博士

【摘要】 分布式线控底盘通过融合驱动、转向、制动以及悬架线控技术,相较于传统车辆,具有更广泛的可控性与更高的灵活性,使其成为自动驾驶的理想平台。在高阶自动驾驶条件下,自动驾驶车辆将全面脱离对人类驾驶员监测与干预的依赖,需要自动驾驶车辆具备全域高性能、安全、稳定行驶能力。而紧急避障场景下的实时轨迹规划与精准运动控制是实现高阶自动驾驶过程中保证行车安全的关键技术。如何将分布式线控底盘的高可控性、多余度特性与自动驾驶技术相结合,提供一个更加高效、安全、智能的载体平台,是目前急需解决的技术难题。本文依托国家自然科学基金面上项目“考虑执行器失效的智能线控底盘运动风险态势评估与一体化运动控制方法研究”(项目编号:52372412)、重庆市技术创新与应用发展专项重点项目“下一代EV新构架智能驾驶仿真平台”(项目编号:CSTB2022TIAD-DEX0014)。针对分布式线控底盘,进行紧急避障时空轨迹规划以及耦合运动控制方法设计,旨在面对动态障碍物的紧急避障场景下,实时规划得到安全避障、满足车辆动力学约束的紧急避障轨迹;利用惯性传感器(Inertial Measurement Unit,IMU)信息以及底盘反馈状态信息实现车辆动力学关键参量准确估计;并充分利用分布式线控底盘的多余度特性,设计高性能的耦合运动控制方法实现对紧急避障轨迹的准确跟踪,在保证车辆稳定性的前提下,充分发挥分布式线控底盘的底盘运动性能。具体研究内容如下:(1)针对存在动态障碍物的紧急避障场景,本文进行分布式线控底盘的时空轨迹规划算法的设计。首先根据障碍物以及周围环境信息,进行三维时空栅格地图的构建,并利用改进混合A*算法进行粗轨迹的搜索。根据分布式线控底盘的运动能力边界,通过加速度的采样、剪枝在保证节点扩展性的同时提高搜索效率,生成对应运动基元。在启发式函数中考虑碰撞风险,减少对碰撞风险较高的子节点的无效探索。同时提出了基于最优边界值问题(Optimal Bundary Value Problem,OBVP)的One-Shot搜索算法,提高目标状态附近搜索效率。在搜索得到的粗轨迹基础上,进行线性安全走廊约束的构建,以保证生成紧急避障轨迹的碰撞安全。最后,利用无侧偏车辆模型作为预测模型,根据安全走廊信息与分布式线控底盘的运动能力边界约束,通过模型预测控制算法优化求解得到紧急避障轨迹。本文在仿真环境下将设计的紧急避障时空轨迹规划算法与EMPlanner算法以及基于非线性模型预测控制的轨迹规划算法进行对比验证。(2)针对高精度车辆状态测量设备成本高且使用场景受限、基于机理模型的车辆状态估计方法存在稳态易偏移、模型易失配等问题,本文设计了一种数据-机理模型混合驱动的车辆动力学关键参量估计方法。考虑到车辆状态数据为典型的时序数据且基于机理模型的状态估计结果中包含丰富的真实状态信息,本文创新性的将基于机理模型的车辆侧向速度估计结果,联合IMU传感器信息、底盘状态反馈信息作为数据驱动方法的输入,并利用长短期记忆神经网络(Long Short-Term Memory,LSTM)结构实现车辆侧向速度准确估计。针对数据驱动估计方法受限于数据集的规模,且存在泛化性与场景适应能力差的问题,本文将侧向速度估计神经网络进行特征提取层与拟合输出层的结构划分。利用5～10分钟的训练样本,对复杂的特征提取层参数进行微调,而着重于使用有限样本对结构相对简单的拟合输出层参数进行重新训练,实现预训练模型在目标域上的快速部署与迁移。同时,本文设计轮胎参数在线自适应模块(Tire Parameter Online Adaptive Module,TPOAM),将估计得到的侧向速度融合估计值作为伪测量值,利用双轨扩展卡尔曼滤波器对车辆前、后轴等效侧偏刚度、车辆状态进行联合估计。最后,在仿真环境下对数据-机理模型混合驱动的车辆动力学关键参量估计效果、基于小样本的迁移学习效果以及TPOAM的侧偏刚度估计效果进行验证。(3)针对分布式线控底盘的高可控性、多余度的特点,本文提出了一种基于分层架构的耦合运动控制方法。在运动跟踪层中,基于变步长离散线性时变模型预测控制器进行目标整车控制量的计算。为了防止分层式控制器的建模误差引起上层生成的控制目标无法被下层控制器实现而导致的分层式控制器控制效果的恶化,本文设计了基于多目标优化的分层控制特性建模方法,并在运动跟踪层的预测模型中进行状态增广。根据上一时刻的预测状态和控制序列信息对线性化参考点进行自适应变化,以提高预测模型精度,同时在预测时域内通过变步长离散方法,在不增加计算负担的情况下有效的扩展预测时域的时间长度。在控制分配层中,考虑垂向载荷转移与各轮摩擦圆约束,利用二次规划算法计算得到各轮目标驱动力与前、后轴目标侧向力。在执行层,本文利用TPOAM输出的在线更新前、后轴等效侧偏刚度值,以准确跟踪控制分配层计算得到的目标前、后轴侧向力。最后在硬件在环(Hardware-in-the-Loop,HIL)台架上对所提出的分布式线控底盘耦合运动控制方法的紧急避障轨迹跟踪效果以及耦合运动控制效果进行验证。(4)搭建HIL台架进行综合工况的仿真验证,并利用分布式线控底盘实验平台进行车辆动力学关键参量估计以及耦合运动控制方法的实车实验验证。利用Speed Goat、NI PXIe-1071仿真器以及工控机搭建联合HIL台架,选取典型的紧急避障场景,对本文提出的分布式线控底盘紧急避障时空轨迹规划与耦合运动控制方法进行综合验证。考虑到仿真以及台架实验与真实的实车环境仍然存在差异,本文利用分布式线控底盘实验平台进行车辆动力学关键参量估计的训练样本采集,并进行数据-机理模型混合驱动的车辆侧向速度估计神经网络的训练,在实车环境下对侧向速度估计以及TPOAM的前、后轴等效侧偏刚度的估计效果进行验证。为了在实车环境下实现神经网络的快速部署,本文进行了从仿真环境到实车环境、以及实车环境下从四轮转向模式到两轮转向模式下的小样本车辆侧向速度估计神经网络的迁移学习验证。最后,本文利用分布式线控底盘实验平台进行耦合运动控制对比验证以及实验场场景综合验证。实验结果表明本文提出的耦合运动控制方法在保证路径以及速度跟踪精度的同时,有效的将车辆状态控制在整车摩擦圆的边界,充分发挥分布式线控底盘的运动潜力。更多还原

【Abstract】 The distributed X-by-wire chassis has higher controllability and flexibility compared to traditional vehicles by integrating drive,steering,braking,and suspension X-by-wire technologies,making it an ideal platform for autonomous driving.Under advanced autonomous driving conditions,autonomous vehicles will completely break away from the dependence on human driver monitoring and intervention,and require autonomous vehicles to have comprehensive,high-performance,safe,and stable driving capabilities.Real-time trajectory planning and precise motion control in emergency obstacle avoidance scenarios are key technologies for ensuring driving safety in achieving high-order autonomous driving processes.How to combine autonomous driving technology with the redundancy characteristics of distributed X-by-wire chassis to provide a more efficient,safe,and intelligent platform for autonomous driving is currently an urgent technical problem that needs to be solved.This study is supported in part by the National Natural Science Foundation of China “Research on Risk Situation Assessment and Integrated Motion Control Method for Intelligent Wire Controlled Chassis Motion Considering Actuator Failure”(Grant No.52372412),the Technology Innovation and Application Development Special Project of Chongqing “Intelligent Driving Simulation Platform for Next-Generation EVs with New Architecture”(Grant No.CSTB2022TIAD-DEX0014).For distributed X-by-wire chassis,emergency obstacle avoidance trajectory planning and coupled motion control methods are designed.The aim of this study is to plan the emergency obstacle avoidance trajectory that safely avoids obstacles and meets the vehicle dynamics constraints in real time,and utilize the output of IMU sensor and the chassis feedback state information to realize the accurate estimation of the key vehicle dynamic parameters.To make full use of the redundancy characteristics of the distributed X-by-wire chassis,it is necessary to design a high-performance coupled motion controller to realize the accurate tracking of emergency obstacle avoidance trajectories,and to give full play to the chassis motion performance of distributed vehicles under the premise of ensuring the vehicle stability.The research content of this study mainly includes the following aspects:(1)For emergency obstacle avoidance scenarios with dynamic obstacles,this study carries out the design of spatiotemporal trajectory planning algorithm.Firstly,according to the obstacles and the surrounding environment information,the construction of 3D spatiotemporal grid map is carried out,and the improved hybrid A* algorithm is utilized for the search of coarse trajectory.According to the boundary of motion capability of distributed X-by-wire chassis,the motion primitives are generated by vehicle acceleration sampling and pruning to improve the search efficiency while ensuring the node scalability.The collision risk is considered in the heuristic function to minimize the ineffective exploration of child nodes with high collision risk.Meanwhile,the One-Shot search algorithm based on the OBVP is designed to improve the search efficiency near the target state.Based on the coarse trajectory obtained from the improved hybrid A* search,the construction of linear safety corridor constraints is achieved to ensure the collision safety of the generated emergency obstacle avoidance trajectory.Using the vehicle model without sideslip as the prediction model,based on the information of the safety corridor and the boundary of motion capability of distributed X-by-wire chassis,the emergency obstacle avoidance trajectory is obtained through the model prediction control algorithm.Finally,this study compares the designed emergency obstacle avoidance trajectory planning algorithm with the EMPlanner algorithm and the trajectory planning algorithm based on nonlinear model predictive control for verification.(2)Aiming at the problems of high cost and limited usage scenarios of high-precision vehicle state measurement equipment,and the problems of steady-state offset and model mismatch in the mechanism model-based vehicle state estimation method,this study designs a fusion estimation method of the key vehicle dynamic parameters driven by a hybrid data-mechanism model.Considering that the vehicle state is typical time series data and the state estimation results based on the mechanism model contain rich information aligned with the actual vehicle state,this study innovatively combines the estimation results of the mechanism model based methods with IMU sensor state detection information and feedback state information of the chassis domain as inputs for data-driven methods,and uses a long short-term memory neural network structure to achieve vehicle lateral velocity estimation.In response to the limitations of data-driven estimation methods on the size of the dataset and the problems of poor generalization and scene adaptability,this study divides the lateral velocity estimation neural network into two parts: feature extraction layer and regression output layer.The parameters of the complex feature extraction layer are fine-tuned,while the focus is on retraining the parameters of the relatively simple structure of the regression output to realize the rapid deployment and migration of the pre-trained model on the target domain using 5～10 minutes training samples.At the same time,this study designs a TPOAM,which uses the estimated lateral velocity as a pseudo measurement value.The dual extended Kalman filter is used to jointly estimate the cornering stiffnesses and vehicle states.Finally,the estimation performance of the key vehicle dynamic parameters driven by the hybrid data-mechanism model proposed,transfer learning based on small samples,and estimation performance of TPOAM module were verified in a simulation environment.(3)For high controllability and redundancy of distributed X-by-wire chassis,this study proposes a hierarchical coupling motion control method.In the motion-following layer,the target global force/moment is calculated based on VTDLTV-MPC.To prevent the deterioration of the control performance caused by the modeling error of the hierarchical controller that the control objectives generated in the upper layer cannot be realized by the lower layer controller,this study designs a hierarchical control characteristic modeling method based on multi-objective optimization with state augmentation and expansion of the prediction model in the motion-following layer.The linearization reference point is updated with the predicted state and control sequence of the last moment to improve the prediction model accuracy,and the time length of the prediction time domain is effectively extended without increasing the computational burden by the variable-time steps discretization method in the prediction domain.In the control-allocation layer,considering the vertical load transfer and the friction circle constraint of each wheel,the quadratic programming algorithm is used to calculate the target driving force of each wheel and the target lateral force of the front and rear axle.In the execution layer,this study utilizes the online updated front and rear axle equivalent cornering stiffnesses from the TPOAM to accurately track the target front and rear axle lateral forces calculated in the control-allocation layer.Finally,the emergency obstacle avoidance trajectory tracking performance,and coupled motion control performance of the proposed method are verified on the HIL platform.(4)The HIL experimental platform for the simulation verification of comprehensive working conditions is built,and utilize the distributed X-by-wire chassis experimental platform to carry out the verification of the estimation of the key dynamic parameters as well as the control performance of the proposed coupled motion controller in real-vehicle experiments.Using Speed Goat,NI PXIe-1071 simulator and IPC to build a joint simulation hardware-in-the-loop experimental platform,the spatiotemporal obstacle avoidance trajectory planning and coupled motion control methods proposed in this study is comprehensively validated in the typical emergency obstacle avoidance scenarios.Considering that there are still differences between the simulation and the real vehicle environment,this study utilizes the distributed X-by-wire chassis experimental platform to collect training samples for the estimation of key parameters of vehicle dynamics,and conducts the training of the vehicle lateral velocity estimation neural network.The effectiveness of the lateral velocity estimation and the estimation of the equivalent cornering stiffnesses of the front and rear axles of the TPOAM is validated in a real vehicle environment.To achieve rapid deployment of neural networks in real vehicle environments,this study conducted transfer learning validation of vehicle lateral velocity estimation neural networks from simulation to real vehicle and from four-wheel steering mode to two-wheel steering mode in real vehicle environments.Finally,the distributed X-by-wire chassis experimental platform is utilized to carry out comparative validation of coupled motion control and validation of the experimental field scenario.The experimental results show that the proposed coupled motion control method effectively controls the vehicle state at the boundary of the friction circle while ensuring the accuracy of path and speed tracking,giving full play to the motion potential of the distributed X-by-wire chassis.更多还原