露天矿作业区无人矿车协同通行决策方法研究

露天矿无人矿车在装卸载作业区内运输过程中的长时间停车等待是制约露天矿无人运输系统效率提升的瓶颈。为提高无人矿车的运输效率,本文结合作业区内的运输作业流程,提出一种基于动态可行驶距离的多车协同通行决策方法。首先,将决策模型建模为混合整数线性规划(Mixed Integer Linear Programming, MILP)模型,表述优化目标和问题约束;其次,考虑到求解MILP模型存在难以满足动态决策实时性的问题,基于蒙特卡洛树搜索(Monte Carlo Tree Search,MCTS)实现多车冲突消解,核心思想是利用搜索树的推演能力进行多车通行前瞻模拟,计算多车的最优通行优先级,动态调整多车的可行驶距离;此外,根据无人矿车在作业区内的作业特征设计不同的MCTS节点价值函数,实现综合考虑运输效率与作业特征的通行优先级排序;最后,设计作业区4,8,12个停车位场景下的多车通行仿真实验,与基于先到先服务(First-Come-FirstServed, FCFS)的方法进行对比,吞吐量提升22.03%~28.00%,平均停车等待时间缩短31.71%~50.79%。同时,搭建微缩智能车辆的6停车位作业区场景实验平台,多车单次运输作业总用时相比FCFS缩短了18.84%。仿真与微缩智能车辆的实验结果表明,本文提出的方法能够提升露天矿作业区多车运输效率。

基于DDPG的无人车智能避障方法研究

本文中提出一种基于强化学习的无人车智能避障方法。鉴于无人车运动必须满足内外约束,包括汽车动力学约束和交通规则约束,且动作输出必须连续,而传统强化学习无法应对连续动作空间问题,提出了一种改进的DDPG算法,解决连续动作空间问题,实现转向盘转角和加速度的连续输出;采取多源传感器数据融合,满足无人车避障算法的状态输入;增加车辆内外约束条件,使输出动作更合理有效。最后,在开源仿真平台TORCS进行仿真,验证了算法的有效性和鲁棒性。

基于神经网络的车辆强制换道预测模型

针对高速公路上车辆行驶速度快,换道行为危险程度高的问题,聚焦于不可避免、发生频繁、一旦发生事故后果严重的强制换道行为,改进基于门控循环单元(GRU)的换道模型,对强制换道行为进行分析与预测。为保证模型的有效性,选取下一代仿真技术(NGSIM)数据作为模型的训练集与检测集,使用侧向加速度将车辆侧向摆动数据有效删除并得到强制换道的最迟换道点,进而实现车辆位置与换道决策的预测。实验结果证明,所提模型能够以96.01%的准确率判定车辆在最迟换道点的强制换道行为,相较于LSTM模型准确率提升了3.67%,同时相较于朴素贝叶斯网络准确率提高了7.31%。

露天矿区无人驾驶行车风险评估及防控策略仿真研究

行车风险评估及防控是露天矿区无人运输系统的关键技术之一,为保障无人驾驶车辆在露天矿区安全运行,从露天矿区车-路-云的运输场景出发,建立了行车安全保障模型。综合车端-路侧-云平台的多源信息,对无人驾驶车辆进行行车风险等级评估并设计了相应的行车风险防控策略。行车安全保障模型由行车状态感知模块、行车风险评估模块和行车风险防控模块3部分组成。在行车风险评估方面,引入车辆前方道路坡度信息对预碰撞时间指标的阈值进行修正,引入车辆前方道路坡度及车辆载重状态信息对最小制动安全距离指标进行修正,结合预碰撞时间指标和最小制动安全距离指标提出了综合行车风险评估策略,可以对露天矿区无人驾驶车辆实时行车碰撞风险进行等级量化。然后,基于有限状态机设计了考虑不同行车风险等级的碰撞风险防控决策系统,针对不同状态制定了满足最小安全距离要求的车辆平稳制动控制策略。最后,基于PreScan与Matlab联合仿真技术,搭建了某露天矿区无人驾驶车辆的数字孪生仿真系统,并进行了水平道路场景、上下坡道场景和满载工况的车辆遇障停车仿真测试。仿真结果表明:行车风险评估模块在下坡道场景可以提前评估风险并及时制动停车,说明引入前方道路坡度信息的综合行车风险评估策略可以提高上下坡道场景车辆的行车安全;同时,通过引入车辆载重信息来修正最小制动安全距离指标,可及时评估满载车辆前方的潜在碰撞风险,提高了露天矿区满载车辆的行车安全;行车风险防控模块的紧急制动控制策略可以在10 m安全距离前实现平稳制动停车,提高了大载重车辆在遇障停车时的平稳性。

面向非结构化道路场景的车辆全局速度规划

为实现智能网联车辆在复杂非结构化道路场景下行驶速度规划,该文提出了一种基于分段匀加速模型的车辆全局速度规划方法。以驾驶安全平稳性为原则分析车辆动力学特性,计算侧翻与侧滑临界车速作为非结构道路上各路径点的极限行驶速度;设计分段匀加速模型建立全局速度规划问题,构建考虑效率、平稳性及能耗的综合损失函数,考虑非结构化道路的坡度曲率连续变化,设计车辆速度、加速度及加加速度变量边界来约束决策变量;结合重型电动车辆制动能量回收功能,针对性提出非结构化道路场景下电动车辆速度规划模型,并通过仿真对方法进行验证。结果表明:主车的加速度范围稳定在-1.0~1.0 m/s^(2),加加速度范围稳定在-0.5~0.8 m/s^(3);相比于基于动态规划的速度规划,所提出的速度规划算法保证车辆行驶平稳性的同时,减少了车辆控制输入,0.8 s的计算时间可满足无人驾驶车载设备实时计算需求;将提出的算法应用于无人矿卡实车,车辆行驶平稳,最大加加速度不超过0.45m/s^(3)。

考虑临近车道行人对交通流影响的改进跟驰模型

基于实际人车混合行驶的情形,提出了一个改进的车辆动力学模型。该模型考虑了邻车道或横向的行人、自行车等对主干道车辆的驾驶行为影响。基于经典的最优速度模型,将主干道车辆与行人之间的横向距离和纵向距离作为参数引入最优速度模型中。为了验证所提模型的稳定性和有效性,利用线性稳定性理论,推导出模型的稳定性条件和非稳定性条件,绘制中性稳定性曲线,直观描述了交通流稳定区域大小。理论结果表明:考虑横向行人干扰因素的改进跟驰模型比传统的只考虑单一车道车辆因素的跟驰模型更加的稳定,并且不同参数的变化,所引起的稳态区域也会发生变化。采用了更加真实的优化速度方程,通过仿真实验来描述车辆的驾驶行为;仿真实验列举了2种实际场景:行人稀少和行人较多时。分别绘制了车辆的速度-时间变化曲线以及车辆的时空图,实验结果表明:横向行人的确会干扰车辆的正常驾驶;不同场景下,行人数量的多寡也会对车辆的驾驶行为造成不同程度的影响。

Development of a Low-cost Hardware-in-the-loop Simulation System as a Test Bench for Anti-lock Braking System

Nowadays validation of anti-lock braking systems（ABS） relies mainly on a large amount of road tests.An alternative means with higher efficiency is employing the hardware-in-the-loop simulation（HILS） system to substitute part of road tests for designing,testing,and tuning electronic control units（ECUs） of ABS.Most HILS systems for ABS use expensive digital signal processor hardware and special purpose software,and some fail-safe functions with regard to wheel speeds cannot be evaluated since artificial wheel speed signals are usually provided.In this paper,a low-cost ABS HILS test bench is developed and used for validating the anti-lock braking performance and tuning control parameters of ABS controllers.Another important merit of the proposed test bench is that it can comprehensively evaluate the fail-safe functions with regard to wheel speed signals since real tone rings and sensors are integrated in the bench.A 5-DOF vehicle model with consideration of longitudinal load transfer is used to calculate tire forces,wheel speeds and vehicle speed.Each of the four real-time wheel speed signal generators consists of a servo motor plus a ring gear,which has sufficient dynamic response ability to emulate the rapid changes of the wheel speeds under strict braking conditions of very slippery roads.The simulation of braking tests under different road adhesion coefficients using the HILS test bench is run,and results show that it can evaluate the anti-lock braking performance of ABS and partly the fail-safe functions.This HILS system can also be used in such applications as durability test,benchmarking and comparison between different ECUs.The test bench developed not only has a relatively low cost,but also can be used to validate the wheel speed-related ECU design and all its fail-safe functions,and a rapid testing and proving platform with a high efficiency for research and development of the automotive ABS is therefore provided.

An Improved Cooperative Adaptive Cruise Control(CACC) Algorithm Considering Invalid Communication

For the Cooperative Adaptive Cruise Control （CACC） Algorithm, existing research studies mainly focus on how inter-vehicle communication can be used to develop CACC controller, the influence of the communication delays and lags of the actuators to the string stability. However, whether the string stability can be guaranteed when inter-vehicle communication is invalid partially has hardly been considered. This paper presents an improved CACC algorithm based on the sliding mode control theory and analyses the range of CACC controller parameters to maintain string stability. A dynamic model of vehicle spacing deviation in a platoon is then established, and the string stability conditions under improved CACC are analyzed. Unlike the traditional CACC algorithms, the proposed algorithm can ensure the functionality of the CACC system even if inter-vehicle communication is partially invalid. Finally, this paper establishes a platoon of five vehicles to simulate the improved CACC algorithm in MATLAB/Simulink, and the simulation results demonstrate that the improved CACC algorithm can maintain the string stability of a CACC platoon through adjusting the controller parameters and enlarging the spacing to prevent accidents. With guaranteed string stability, the proposed CACC algorithm can prevent oscillation of vehicle spacing and reduce chain collision accidents under real-world circumstances. This research proposes an improved CACC algorithm, which can guarantee the string stability when inter-vehicle communication is invalid.

产教研融合建设研究生人工智能系列课程探索——以自动驾驶方向人工智能课程为例

针对新成立的自动驾驶专业硕士方向,基于科教融合和产教融合,探索建设了三门人工智能理论及其在自动驾驶领域应用系列课程。基于团队科研成果和企业工程项目凝练项目式教学案例、开发自动驾驶实验教学设备,展开项目研究式教学;基于百度公司产业级人工智能平台完成课程线上教学资源开发,展开线上线下混合式教学;基于自动驾驶新兴产业对人才的需求设计课程思政案例,知识传授中融合价值引领;为我国自动驾驶新兴产业发展培养一流的高层次紧缺人才和卓越工程师做出贡献,也可为其他学科人工智能课程的建设提供参考和借鉴。

Vehicle's fuel consumption of car-following models

Many car-following models with their own merits have been proposed to explore various traffic problems,but researchers have rarely used vehicle's performance to evaluate them.In this paper,we study each vehicle's fuel consumption of the optimal velocity model,full velocity difference model,full velocity and acceleration difference model and the car-following model with consideration of the traffic interruption probability under two traffic situations,respectively.The numerical results show that the car-following model with consideration of traffic interruption probability can reduce vehicle's fuel consumption in the two studied traffic situations and thus improve the vehicle's fuel economy.