华南理工大学学报(自然科学版) ›› 2016, Vol. 44 ›› Issue (12): 74-80.doi: 10.3969/j.issn.1000-565X.2016.12.011

• 交通与运输工程 • 上一篇    下一篇

基于驾驶员行为的神经网络无人驾驶控制

张文明 韩泓冰 杨珏 易筱   

  1. 北京科技大学 机械工程学院,北京 100083
  • 收稿日期:2016-05-12 修回日期:2016-08-09 出版日期:2016-12-25 发布日期:2016-11-01
  • 通信作者: 杨珏(1975-),男,博士,副教授,主要从事非公路车辆设计研究. E-mail:yangjue@ustb.edu.cn
  • 作者简介:张文明(1955-),男,博士,教授,主要从事非公路车辆设计、非公路车辆状态检测与故障诊断研究. E-mail:wmzhang@ ustb. edu. cn
  • 基金资助:

    国家高技术研究发展计划(863 计划)项目(2011AA060404)

A Neural Network-Based Autonomous Articulated Vehicle System Considering Driver Behavior

ZHANG Wen-ming HAN Hong-bing YANG Jue YI Xiao   

  1. School of Mechanical Engineering,University of Science and Technology Beijing,Beijing 100083,China
  • Received:2016-05-12 Revised:2016-08-09 Online:2016-12-25 Published:2016-11-01
  • Contact: 杨珏(1975-),男,博士,副教授,主要从事非公路车辆设计研究. E-mail:yangjue@ustb.edu.cn
  • About author:张文明(1955-),男,博士,教授,主要从事非公路车辆设计、非公路车辆状态检测与故障诊断研究. E-mail:wmzhang@ ustb. edu. cn
  • Supported by:

    Supported by the National High-Tech R&D Program of China(863 Program)(2011AA060404)

摘要: 针对铰接式自卸车的转向特性,提出了一种基于驾驶员行为的神经网络无人驾驶控制方法. 建立了以激光雷达、角度传感器为主要环境信息的采集系统,通过分析铰接
式车辆转向特征建立铰接式自卸车运动学模型和动力学模型,利用 ADAMS 动力学软件建立车辆动力学模型并进行车辆稳态测试. 建立基于最优预瞄控制的人工神经网络控制算法的驾驶员模型,通过 ADAMS-Matlab/Simulink 联合仿真验证模型. 最后搭建真实巷道环境进行直线路段回正实验和曲线路径跟踪实验,结果显示,该控制模型在变曲率路段中,横向位置偏差小于可通过路径宽度的 10%,航向角偏差优化 90%,表明该神经网络驾驶员控制模型收敛速度快,稳态特性好,具有良好的无人驾驶能力.

关键词: 驾驶员模型, 神经网络, 动力学模型, 联合仿真

Abstract:

In view of the steering characteristics of articulated dump trucks,an autonomous articulated vehicle sys- tem is proposed based on neural networks and by considering driver behaviors.First,a sensor collecting system based on the laser radar and the angular transducer is established,and an articulated vehicle kinematics model and a dynamics model of articulated dump trucks are constructed by analyzing the steering characteristics of articulated dump trucks.Then,by using the ADAMS software,a dynamic model of the trucks is constructed to perform a steady state test.Moreover,a driver model of the artificial neural network control algorithm is constructed based on the optimal preview control,and it is verified by an Adams-Matlab/Simulink co-simulation.Finally,this control model is also verified by establishing a simulation ground tunnel to perform the straight-road-return and curve-road- following tests.The results show that,when the constructed control model is applied to the variable curvature road,the lateral position error is less than 10% of the passable distance,and 90% of the course angle deviation is opti- mized,which indicates that the constructed control model has a high convergence speed,a good steady state and an excellent unmanned driving performance,

Key words: driver model, neural networks, dynamic model, co-simulation

中图分类号: