面向前车的驾驶行为感知与意图识别算法研究

感知周围车辆的驾驶行为并识别其意图将成为新一代高级驾驶辅助系统的重要组成部分。针对现有方法只考虑单一驾驶行为且可扩展性和可伸缩性差,提出一种基于稀疏表示理论的驾驶行为感知字典模型(Driving Behavior Perception Dictionary Model, DBPDM)。将车辆行驶状态视为时间序列,设计基于自回归积分移动平均(Autoregressive Integrated Moving Average,ARIMA)结合在线梯度下降(Online Gradient Descent, OGD)优化器的在线预测模型,提出基于驾驶行为预测的意图识别构架(Intention Recognition Framework, IRF)。首先,采用图Lasso方法估计典型驾驶行为的稀疏逆协方差矩阵构建驾驶行为字典库,并采用Logdet散度方法计算各逆协方差矩阵的差异获得行为感知字典模型。然后,基于在线预测模型对目标车辆的行驶轨迹和运动状态进行预测,结合主车车辆的行驶状态作为稀疏表示的观测信号,以获取预测时域内的目标车辆意图。最后,采用NGSIM (Next Generation SIMulation)真实驾驶数据对模型进行开发和测试。研究结果表明:所提出的行为感知模型能对6种典型驾驶行为构建行为字典,在分类准确率上与现有方法相比有明显提升,对换道和转向行为样本的平均识别准确率分别达到99.1%和92.9%;该模型能够在相对早期阶段准确地识别出车辆行为;在线预测算法能较好预测出目标车辆的行驶轨迹和运动状态,从而间接地反映出其在预测时域内的驾驶意图;IRF可在换道和转向行为开始前的1.5 s较为准确地识别出目标车辆的意图,平均识别准确率超过80%。     

商用车电液复合转向系统的车道保持策略

为了提高商用车的行驶安全性,避免因驾驶人的分心驾驶出现车辆偏离车道的问题,提出一种基于电液复合转向系统的商用车车道保持策略;在建立电液复合转向系统模型、二自由度车辆模型、预瞄驾驶人模型的基础上,设计基于驾驶人在环的MPC和ADRC串级的车道保持控制策略。首先,采用MPC算法将车辆横向位置控制的最优问题转化为二次规划求得目标前轮转角;然后,考虑电液复合转向系统的不确定和干扰问题,利用ADRC算法对目标转向盘转角和实际驾驶人的转向盘转角差值以转矩信号的形式进行补偿。同时研究车道保持系统对驾驶人的干预问题,引入干预系数的概念,采用模糊控制的方法,将驾驶人手力和车辆的运动状态作为输入变量,干预系数作为输出变量,保证整车行驶安全性的前提下减小车道保持辅助系统对驾驶人的干预。最后,通过MATLAB/Simulink仿真和硬件在环试验对所设计的控制策略进行验证。研究结果表明：所设计的基于商用车电液复合转向系统的车道保持策略能够及时地纠正因驾驶人的分心驾驶而导致车辆偏离所在行驶车道的行为,特别是在弯道处出现驾驶人转向不足或过度转向的情况时,能够将车辆维持在车道线之内,保证车辆的行驶安全性,同时由于干预系数的设计,使得驾驶人也有良好的人机交互体验感。     

多车道复杂环境下驾驶倾向性状态辨识

驾驶倾向性是汽车行驶过程中操控者情感偏好等特征的动态测度,是车辆安全驾驶辅助系统,特别是其碰撞预警系统中汽车驾驶人意图等心理、意识计算必须考虑的核心参数.以3车道场景为例,分析目标车位于不同车道时周边的交通态势(主要指车辆集群编组关系,重点以目标车位于中间车道为例),设计实验采集人、车、环境等相关动态信息,获取不同态势下驾驶倾向性特征数据,利用动态贝叶斯网络建立时变环境下驾驶倾向性动态辨识模型.实验验证表明,所建模型对驾驶人倾向性类型的辨识准确率可达到91％以上,能够适应多车道情况下驾驶人倾向性类型的动态识别,为以人为中心的个性化汽车主动安全系统的实现提供理论基础.     

针对自适应巡航的驾驶人风格辨识

车辆进入自适应巡航工况下行驶时,不同风格的驾驶人会对自适应巡航控制系统(Adaptive Cruise Control,ACC)有不同的需求。文章首先通过对不同驾驶人在9种跟随试验下获取的实验数据分析,选取表征驾驶人风格的驾驶特征参数;其次对所有驾驶人驾驶特征参数利用K-mean算法聚类分析,将驾驶人三类,并利用BP神经网络建立辨识模型对驾驶人风格进行辨识。结果表明;文章提出的方法可以较高的准确率对驾驶人风格进行分类,提高自适应巡航系统适应驾驶人的能力。     

基于 AHP 和物元分析的愤怒驾驶状态辨识研究

驾驶人在愤怒情绪下的驾驶行为是影响车辆行驶安全性的重要因素之一,愤怒驾驶情绪的产生及其程度受到驾驶人自身和道路交通环境中多因素的影响。文中综合考虑驾驶人自身因素和行车环境对驾驶状态的影响,提出了愤怒驾驶状态的辨识方法。文中筛选了与愤怒驾驶行为相关的驾驶人因素和道路环境因素,构建了1个驾驶人愤怒状态辨识的层次分析模型,并根据相关因素之间对愤怒驾驶行为影响的重要程度构造判断矩阵,求出各相关因素对愤怒驾驶行为的影响权值。应用综合权重的物元多属性决策方法辨识驾驶人的愤怒驾驶状态及程度。应用所提出的方法对22组实车试验中出现的愤怒驾驶状态进行辨识,结果表明,72.7%的结果与实车实验所得的结果相符,因此,该方法可对愤怒驾驶行为进行识别。文中所提出的方法能够融合驾驶人因素和环境因素对愤怒驾驶行为的影响,有效的辨识出驾驶人的愤怒驾驶状况及程度。      

基于机器视觉的客运车辆危险行驶状态辨识技术研究

为了提高客运车辆危险驾驶状态的辨识效能,提出利用机器视觉辨识危险驾驶行为。使用车载CCD实时采集路面图像,依据图像处理算法检测车道标识线的位置以及本车与前方目标车辆的实时距离,建立了车辆横向偏航辨识模型和车辆纵向危险行驶状态辨识模型;结合所建模型的辨识结果,确定预警方案。实车试验结果表明,所建模型能有效辨识车辆横向偏航和纵向跟车过近危险行驶行为,可用来降低潜在的危险驾驶行为,提高车辆在途行驶安全性。     

险情中驾驶人接管自动驾驶车辆的驾驶行为研究

为了指导驾驶人安全地使用自动驾驶系统,在模拟驾驶仿真平台上研究驾驶人在复杂交通环境下实现自动驾驶切换到手动驾驶的绩效表现,以及影响驾驶人驾驶方式切换过程的因素。试验收集了36名驾驶人在2种险情下切换驾驶方式过程中的驾驶行为数据,包括驾驶速度、车头间距、车辆横向位移和车辆转向等数据。研究结果表明:部分驾驶人会在险情提示出现之后即刻切换驾驶方式,而部分驾驶人则在提示出现之后继续监控道路和车辆信息,直到交通冲突出现的时候才进行驾驶方式切换;对上述2种切换方式进行独立分析发现,进行第2种切换的驾驶人在接管车辆之前所处的状态(玩游戏或者听音乐)对驾驶人接管车辆所用的切换时间有显著性影响,驾驶人处于玩游戏的状态下接管车辆所用的时间会小于驾驶人在听音乐状态下接管车辆所用的时间;切换方式对车辆横向位置有显著性影响,而且进行第1种切换的驾驶人在后续的驾驶过程中能保持较小的安全速度行驶,说明"切换方式1"与"较小的安全车速"存在相关关系,但二者的因果关系尚需深一步研究。     

车车通信环境下基于驾驶意图共享的车辆避撞预警算法

针对传统基于距离或时间的车辆避撞预警算法存在较高误警率的问题，考虑在避撞预警算法中引入驾驶意图共享的概念，提出了基于实际外场复杂车车通信V2V（Vehicle-to-vehicle，V2V）环境下的车辆跟驰避撞预警算法。基于LTE-V构建外场V2V环境，将车辆行驶过程描述为一个时间序列的隐性马尔可夫随机过程，借助隐马尔可夫模型（Hidden Markov Model，HMM）建立驾驶人驾驶意图与车辆相对行驶状态序列之间的隐含关系模型，并给出基于Viterbi算法的驾驶意图预判求解方法，将驾驶意图作为特征因子集成到安全距离模型中，提出基于驾驶意图共享的避撞（Driving Intention Based Collision Avoidance，DI-CA）预警算法。利用构建的V2V试验环境，实现了匀速、加速、减速和紧急制动等4种驾驶意图，以及相对速度和相对距离增加、减小、保持不变等9种组合的车辆行驶状态试验数据获取，并利用试验数据对所提出的DI-CA预警算法进行实证分析。结果表明：所提出预警算法能够针对不同驾驶意图提供有效的车辆碰撞预警。在此基础上将4种驾驶意图下的DI-CA预警算法与Mazda预警算法求得的安全距离进行了对比分析，所提出的DI-CA预警算法的平均预警正确率为84%，高于Mazda预警算法的78%，而DI-CA预警算法的平均误警率和漏警率分别为5%和16%，均明显低于Mazda预警算法，说明所提出的DI-CA预警算法在提升预警效果的同时明显降低了误警率和漏警率，可避免行驶过程中因误警而导致的连续刹车，以及因漏警而导致的可能碰撞事故发生。最后，总结并给出了驾驶意图共享理论应用于车辆避撞预警的研究展望。     

融合预瞄与势场栅格法的紧急避撞驾驶人模型

紧急避障工况下的驾驶人操作具有响应快且动作幅值较大的特点，传统预瞄驾驶人模型已不能适应紧急避障工况的需求，故考虑实际避撞场景开发相应的驾驶人模型就显得尤为必要。针对此种状况，基于驾驶模拟器，结合紧急避撞工况实际驾驶人操纵数据，提出了一种融合预瞄与势场栅格法的紧急避撞驾驶人模型。首先针对紧急避撞工况下车辆运动特点，建立车辆横、纵向耦合非线性动力学模型，并给出其状态空间方程描述；其次，离线仿真分析紧急避撞系统特征，并结合线性二次型最优控制，建立最优曲率预瞄+跟踪误差反馈驾驶人模型；再者，基于紧急避撞工况下真实驾驶人经验转向行为数据，开发基于势场栅格法的驾驶人模型，为进一步提高驾驶人模型对避障行驶工况的适应性，将基于势场栅格法的驾驶人模型与最优曲率预瞄+跟踪误差反馈驾驶人模型进行融合，并基于Sigmoid函数实现两者输出的权重分配；最后，针对所提出的融合预瞄与势场栅格法的驾驶人模型，开展基于避撞台架的驾驶人在环仿真试验以及实车试验。研究结果表明：在紧急避撞工况下，对比最优曲率预瞄+跟踪误差反馈驾驶人模型，融合预瞄与势场栅格法的驾驶人模型输出的转向动作与实际驾驶人行为较为接近，可在保证避障安全性的前提下，兼顾避障路径跟踪精度与车辆行驶的稳定性。     

动力模块车组的驾驶真实感研究

为了使驾驶人在驾驶大型动力模块车组时获得力反馈真实感,提出基于伺服电机扭矩控制的转向盘力觉真实感系统.基于普通车辆的转向系特性,建立转弯时单个转向轮的转向力修正模型,据此利用动力模块车组同一轴线上2个悬挂转角与转向盘转角的数学关系,建立动力模块车组转向盘转向力模型,在此基础上构建伺服电机驱动的转向盘转向力模拟试验平台和虚拟车辆行驶环境.结果表明:数字转向盘的转向力矩曲线与普通车辆转向力矩特性相近,符合中国标准的规定,验证了所提方法的有效性.     

基于蚁狮算法的UKF车辆状态参数估计器

为了准确获取分布式驱动电动汽车状态参数信息，满足车辆稳定性控制系统的需求，提出一种基于蚁狮算法的无迹卡尔曼滤波状态参数估计器。针对无迹卡尔曼滤波（UKF）过程中噪声协方差矩阵的不确定性，采用蚁狮优化算法（ALO）对其进行寻优，并引入奇异值分解（SVD）的方法来维持噪声协方差矩阵的正定性，此外，基于指数加权最小二乘法对车辆侧偏刚度进行辨识并将其作为状态参数估计器输入。基于MATLAB/Simulink和CarSim联合仿真平台，建立分布式驱动电动汽车参数估计模型，分别进行双移线工况和正弦迟滞工况仿真，并基于A&D5435快速原型开发平台进行双移线工况实车试验。仿真与试验结果表明：相比于SVDUKF算法估计结果，双移线仿真工况下，基于ALO-SVDUKF算法估计得到的质心侧偏角和横摆角速度的均方根误差分别降低了55.7%、30.7%，正弦迟滞仿真工况下，均方根误差分别降低了58.1%、85.1%，且在车辆处于极限失稳状态时仍能维持较好的估计效果；双移线试验工况下，横摆角速度的估计值与实际测量值之间的均方根误差仅为0.938 4（°）·s^-1；提出的基于ALO-SVDUKF算法的分布式驱动电动汽车状态参数估计器能够有效提高质心侧偏角和横摆角速度的估计精度，可为车辆稳定性控制提供精确的状态信息。     

Vehicle stability control based on driver’s emergency alignment intention recognition

In this work, the reference model modification strategy for vehicle stability control based on driver's intention recognition under emergent obstacle avoidance situation was proposed. First the conflicts between the driver's emergency alignment (EA) intention and vehicle response characteristics were analyzed in critical emergent obstacle avoidance situation. Second combining steering wheel angle and its speed, the driver's EA intention was recognized. The reference model modification strategy based on steering operation index (SOI) was presented. Then a LQR model following controller with tire cornering stiffness adaption was used to generate direct yaw moment for tracking modified reference yaw rate and reference sideslip angle. Finally based on the four-in-wheel-motor-drive (FIWMD) electric vehicles (EV), double lane change and slalom tests were conducted to compare the results using modified reference model with the results using normal reference model. The experimental tests have proved the effectiveness of the reference model modification strategy based on driver's intention recognition.     

Desired yaw rate and steering control method during cornering for a six-wheeled vehicle

This paper proposes a steering control method based on optimal control theory to improve the maneuverability of a six-wheeled vehicle during cornering. The six-wheeled vehicle is believed to have better performance than a four-wheeled vehicle in terms of its capability for crossing obstacles, off-road maneuvering and fail-safe handling when one or two of the tires are punctured. Although many methods to improve the four-wheeled vehicle’s lateral stability have been studied and developed, there have only been a few studies on the six-wheeled vehicle’s lateral stability. Some studies of the six-wheeled vehicle have been reported recently, but they are related to the desired yaw rate of a four-wheeled vehicle to control the six-wheeled vehicle’s maneuvering during corning. In this paper, the sideslip angle and yaw rate are controlled to improve the maneuverability during cornering by independent control of the steering angles of the six wheels. The desired yaw rate that is suitable for a six-wheeled vehicle is proposed as a control target. In addition, a scaled-down vehicle with six drive motors and six steering motors that can be controlled independently is designed. The performance of the proposed control methods is verified using a full model vehicle simulation and scaled-down vehicle experiment.     

识别驾驶人意图的车道偏离防避有界控制

为了防止车辆偏离车道导致交通事故的发生和避免车道偏离防避系统（Lane Departure Avoidance Systems,LDAS）对驾驶人行为不必要的干预,提出基于中心区操纵特性阈值法和基于D-S（Dempster-Shafer）证据理论的车辆偏离车道驾驶人意图识别准则,并运用CarSim/Simulink联合仿真对比2种识别准则的有效性。建立转向盘角速度为输入的车路模型,设计LDAS滑模转向控制器,基于预瞄点的侧向偏移量和横摆角速度设计LDAS的期望横摆角速度观测器,并与基于道路曲率和预瞄点侧向偏移量的期望横摆角速度的LDAS进行性能对比。运用相平面法确定保证LDAS车辆稳定性的前轮转向角最大值,并基于CarSim/LabVIEW RT硬件在环试验平台验证基于BP神经网络训练获得D-S证据理论的初始概率赋值的驾驶人意图决策算法的有效性。结果表明：所提出的识别准则能够及时识别车辆偏离车道时的驾驶人意图,为LDAS控制器介入赢得了宝贵的时间,所设计的期望横摆角速度观测器具有很好的稳定性,所提出的方法能够有效避免车辆偏离车道。     

基于状态空间采样的高速公路智能网联车辆轨迹动态规划

为实现智能网联车辆在高速公路动态行车环境下的轨迹实时规划，提出一种基于状态空间采样的轨迹动态规划方法。首先，以安全性为原则选取主车当前行驶的理想车道。基于Frenet坐标与笛卡尔坐标的转换关系，建立车辆运动横、纵向解耦的独立积分系统。将高速公路常见的行驶状态分为车道保持与定速巡航、变道以及前车跟随3类，预测主车行驶车道并针对3类行驶状态分别设计轨迹终端的目标配置方法。然后，利用多项式函数生成连接初始配置和目标配置的多条待选轨迹。构建考虑轨迹偏离理想车道程度、始末速度变化、规划周期和轨迹舒适性的综合损失函数，结合速度、加速度、曲率检查来评价各条待选轨迹的成本并进行排序。最后，预测车辆的横、纵向运动轨迹并构建一种胶囊形的车辆虚拟安全边界，通过碰撞检测，确定主车的最优轨迹，设置动态规划触发条件及时更新最优轨迹并避免过度规划浪费资源。研究结果表明：提出的算法能满足高速公路场景的动态规划需求；通过对轨迹规划周期、虚拟安全边界、动态规划时间间隔等关键参数的分析与优化，主车的横摆角速度范围稳定在-0.1~0.15 (°)·s^-1，横向加速度范围稳定在-0.16~0.32 m·s^-2，跟踪参考轨迹的最大误差不超过0.022 m，提出的算法能规划出具有高安全性、稳定性和舒适性的轨迹。     

Optimal Control of Four Wheel Steering Vehicle

This paper derives a method of controlling four wheel steering using optimal control theory. The purpose of control is to minimize the sideslip angle at the center of gravity. The control method feeds forward the steering wheel angle and feeds back the yaw velocity and the sideslip angle to the front and rear wheel angles. Theoretical studies show that the sideslip angle is reduced to zero even in the transient state, and that the understeer characteristic and frequency response can be changed regardless of the vehicle static margin. This Paper also examines various characteristics of the influence of the side force nonlinearities of tires and crosswinds.     

Optimal Control of Four Wheel Steering Vehicle

SUMMARY

This paper derives a method of controlling four wheel steering using optimal control theory. The purpose of control is to minimize the sideslip angle at the center of gravity. The control method feeds forward the steering wheel angle and feeds back the yaw velocity and the sideslip angle to the front and rear wheel angles. Theoretical studies show that the sideslip angle is reduced to zero even in the transient state, and that the understeer characteristic and frequency response can be changed regardless of the vehicle static margin. This Paper also examines various characteristics of the influence of the side force nonlinearities of tires and crosswinds.     

Disturbance Observer-Based Sideslip Angle Control for Improving Cornering Characteristics of In-Wheel Motor Electric Vehicles

In this paper, a robust sideslip angle controller based on the direct yaw moment control (DYC) is proposed for in-wheel motor electric vehicles. Many studies have demonstrated that the DYC is one of the effective methods to improve vehicle maneuverability and stability. Previous approaches to achieve the DYC used differential braking and active steering system. Not only that, the conventional control systems were commonly dependent on the feedback of the yaw rate. In contrast to the traditional control schemes, however, this paper proposes a novel approach based on sideslip angle feedback without controlling the yaw rate. This is mainly because if the vehicle sideslip angle is controlled properly, the intended sideslip angle helps the vehicle to pass through the corner even at high speed. On the other hand, the vehicle may become unstable because of the too large sideslip caused by unexpected yaw disturbances and model uncertainties of time-varying parameters. From this aspect, disturbance observer (DOB) is employed to assure robust performance of the controller. The proposed controller was realized in CarSim model described actual electric vehicle and verified through computer simulations.     

Novel stability control strategy for distributed drive electric vehicle based on driver operation intention

In this paper, a novel direct yaw control method based on driver operation intention for stability control of a distributed drive electric vehicle is proposed. It was discovered that the vehicle loses its stability easily under an emergency steering alignment (EA) problem. An emergent control algorithm is proposed to improve vehicle stability under such a condition. A driver operation intention recognition module is developed to identify the driving conditions. When the vehicle enters into an EA condition, the module can quickly identify it and transfer the control method from normal direct yaw control to emergency control. Two control algorithms are designed. The emergency control algorithm is applied to an EA condition while the adaptive control algorithm is applied to other conditions except the EA condition. Both simulation results and real vehicle results show that: The driver module can accurately identify driving conditions based on driver operation intention. When the vehicle enters into EA condition, the emergent control algorithm can intervene quickly, and it has proven to outperform normal direct yaw control for better stabilization of vehicles.     

基于模糊逻辑的车辆侧偏角估计方法

提出了一种基于模糊逻辑的汽车侧偏角估计方法。它利用模糊逻辑和汽车运动学模型,将汽车转向盘转角、车轮转速、汽车加速度和横摆角速度信息相融合,进行车辆侧偏角估计。试验结果显示,该方法的鲁棒性和精确性较好,而且响应频率较高,可以满足ESP的控制需要。