吉林大学学报(工学版) ›› 2023, Vol. 53 ›› Issue (3): 653-662.doi: 10.13229/j.cnki.jdxbgxb20221273

• 通信与控制工程 • 上一篇    

基于Transformer的轮式里程计误差预测模型

何科(),丁海涛,赖宣淇,许男,郭孔辉   

  1. 吉林大学 汽车工程学院,长春 130022
  • 收稿日期:2022-09-29 出版日期:2023-03-01 发布日期:2023-03-29
  • 作者简介:何科(1995-),男,博士研究生. 研究方向:自动驾驶. E-mail:hk_jlu_goon@163.com
  • 基金资助:
    国家自然科学基金项目(U1864206)

Wheel odometry error prediction model based on transformer

Ke HE(),Hai-tao DING,Xuan-qi LAI,Nan XU,Kong-hui GUO   

  1. College of Automotive Engineering,Jilin University,Changchun 130022,China
  • Received:2022-09-29 Online:2023-03-01 Published:2023-03-29

摘要:

针对利用轮式里程计定位时会产生难以预测和多变误差的问题,提出了使用Transformer神经网络建立轮式里程计误差预测模型,以准确预测变化的里程误差,提高了轮式里程计的定位精度。首先,建立不考虑工况特征和考虑工况特征两种模型。然后,在多种工况下与LSTM模型进行对比实验,结果表明:在常规和挑战性工况下,本文模型相比LSTM模型具有更高的精度、稳定性和可靠性。同时,相比于不考虑工况特征,考虑工况特征能有效提高模型的整体性能。

关键词: 车辆工程, 自动驾驶, 定位, 轮式里程计, 深度学习, Transformer模型

Abstract:

To address the problem of unpredictable and variable errors when using wheel odometry for localization, a wheel odometry error prediction model based on Transformer neural network is developed to accurately predict the odometry error that accumulates and changes as the mileage increases, and to improve the accuracy of localization using wheel odometry under GPS occlusion. First, two models were established without and with the driving condition characteristics, then they were compared with the LSTM model under various driving conditions. The experimental results show that the Transformer-based wheel odometry error prediction model can accurately predict the odometry error with higher accuracy, stability and reliability than the LSTM model under both regular driving conditions and challenging driving conditions where it is difficult to measure the odometry signal accurately. At the same time, compared with the Transformer model without considering the driving condition characteristics, the Transformer model with considering the driving condition characteristics improve the performance in all evaluation indexes, which proves that considering the driving condition characteristics can effectively improve the prediction performance of the model.

Key words: vehicle engineering, autonomous driving, localization, wheel odometry, deep learning, Transformer model

中图分类号: 

  • U469.79

图1

轮式里程计误差预测模型在定位系统中的作用"

图2

Transformer模型架构"

表1

选取的特征"

特征序号描述特征序号描述
1里程17急刹车
2环岛18左右急转弯
3山路19换挡
4乡村道路20打滑
5高速公路21连续左右转
6市中心22短时间加速度快速变化
7市区内23U-turn
8蜿蜒道路24倒车
9住宅小区25漂移
10山谷26碰撞
1127Z字形行驶
12土路28近似直线的行驶
13碎石路29静止
14潮湿道路30停车
15泥路31里程误差
16坑洞

图3

两种Transformer模型结构"

表2

用于Transformer模型训练的IO-VNB数据子集"

序号IO-VNB数据子集序号IO-VNB数据子集
1V-Vta1a14V-Vtb1
2V-Vta215V-Vtb2
3V-Vta816V-Vtb3
4V-Vta1017V-Vtb5
5V-Vta1618V-Vw4
6V-Vta1719V-Vw5
7V-Vta2020V-Vw14b
8V-Vta2121V-Vw14c
9V-Vta2222V-Vfa01
10V-Vta2723V-Vfa02
11V-Vta2824V-Vfb01a
12V-Vta2925V-Vfb01b
13V-Vta3026V-Vfb02b

表3

Transformer模型训练参数"

参数IO-VNB数据子集
编码器和解码器个数N6
子层以及嵌入层的输出维度dmodel64
全连接层中间层层数dff32
注意力头数h8
批量大小32
训练次数Epochs100
时间步长(滑动窗口)/s11
随机失活比例Dropout0
激活函数ReLU

表4

用于性能评估的IO-VNB数据子集"

场景IO-VNB数据子集行驶总时间/min行驶距离/km速度/(km·h-1加速度/(9.8 m·s-2
高速公路V-Vw121.752.6482.6~97.4-0.06~0.07
加速度快速变化V-Vfb02e1.61.5237.4~73.9-0.24~0.19
V-Vta121.11.2744.7~85.3-0.44~0.13
急刹车V-Vw16b2.01.991.3~86.3-0.75~0.29
V-Vw170.50.5431.5~72.7-0.8~0.19
V-Vta90.40.4348.9~87.7-0.6~0.14
湿/泥泞路面V-Vtb81.21.3560.9~76.5-0.35~0.08
V-Vtb110.70.845.1~75.3-0.05~0.12
V-Vtb132.10.997.5~43.3-0.31~0.22

表5

Transformer参数调节过程"

NdmodeldffhDropoutmaxminμσ
基准66432801.990.030.550.48
A66432402.110.030.630.50
664321602.060.020.600.49
B46432802.020.040.620.47
86432802.100.040.730.53
C63232802.000.030.660.49
612832801.970.030.540.46
D66416802.020.040.740.50
66464801.980.020.580.46
E6643280.11.990.030.830.52
6643280.22.000.070.820.46

表6

高速公路工况的测试结果"

指标LSTM模型误差/mTransformer模型误差
不考虑工况特征提升/%考虑工况特征提升/%
max0.680.70-30.5618
min0.250.10600.0580
μ0.460.34260.3035
σ0.080.12-500.15-87.5

图4

高速公路工况的测试结果"

表7

车辆加速度快速变化的工况测试结果"

指标LSTM模型误差/mTransformer模型误差/m
不考虑工况特征提升/%考虑工况特征提升/%
max2.952.00321.9933
min0.540.08850.0591
μ1.401.01280.7249
σ0.790.52340.5530

图5

车辆加速度快速变化的工况测试结果"

表8

急刹车工况的测试结果"

指标LSTM模型误差/mTransformer模型误差
不考虑工况特征提升/%考虑工况特征提升/%
max2.931.97331.6245
min0.100.03700.0370
μ1.220.70430.2381
σ0.840.52380.2373

图6

急刹车工况的测试结果"

表9

湿/泥泞路面工况的测试结果"

指标LSTM模型误差/mTransformer模型误差/m
不考虑工况特征提升/%考虑工况特征提升/%
max2.331.99151.9317
min0.200.2000.0385
μ0.660.6430.5812
σ0.520.44150.4317

图7

湿/泥泞路面工况的测试结果"

图8

原始里程计输出误差与利用误差预测模型补偿后误差的对比"

1 高振海,孙天骏,何磊. 汽车纵向自动驾驶的因果推理型决策[J]. 吉林大学学报: 工学版, 2019, 49(5): 1392-1404.
Gao Zhen-hai, Sun Tian-jun, He Lei. Causal reasoning decision⁃making for vehicle longitudinal automatic driving[J]. Journal of Jilin University (Engineering and Technology Edition), 2019, 49(5): 1392-1404.
2 Jo K, Kim J, Kim D, et al. Development of autonomous car—part I: distributed system architecture and development process[J]. IEEE Transactions on Industrial Electronics, 2014, 61(12): 7131-7140.
3 Lu W, Rodríguez F S A, Seignez E, et al. Lane marking-based vehicle localization using low-cost GPS and open source map[J]. Unmanned Systems, 2015, 3(4): 239-251.
4 Xu L H, Hu S G, Luo Q. A new lane departure warning algorithm considering the driver's behavior characteristics[J]. Mathematical Problems in Engineering, 2015(1): No.412126.
5 Levinson J, Montemerlo M, Thrun S. Map-based precision vehicle localization in urban environments[C]∥Robotics: Science and Systems, Atlanta, Georgia, USA, 2007: No.16.
6 秦晓辉, 王哲文, 庞涛, 等. 基于车辆模型紧耦合的封闭园区车辆定位方法[J]. 汽车工程,2021,43(9): 1328-1335.
Qin Xiao-hui, Wang Zhe-wen, Pang Tao, et al. Vehicle positioning method based on tight coupling of vehicle model in enclosed environments[J]. Automotive Engineering, 2021, 43(9): 1328-1335.
7 Kubo N, Suzuki T. Performance improvement of RTK-GNSS with IMU and vehicle speed sensors in an urban environment[J]. IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences, 2016, 99(1): 217-224.
8 何永明, 陈世升, 冯佳, 等. 基于高精地图的超高速公路虚拟轨道系统[J/OL]. [2022-02-15].
9 Funk N, Alatur N, Deuber R, et al. Autonomous electric race car design[J/OL]. [2022-08-26].
10 Schwesinger U, Bürki M, Timpner J, et al. Automated valet parking and charging for e-mobility[C]∥2016 IEEE Intelligent Vehicles Symposium (IV), Gothenburg, Sweden, 2016: 157-164.
11 Schanz A, Spieker A, Kuhnert K D. Autonomous parking in subterranean garages-a look at the position estimation[C]∥IEEE IV2003 Intelligent Vehicles Symposium, OH, USA, 2003: 253-258.
12 Malleswaran M, Vaidehi V, Manjula S, et al. Performance comparison of HONNs and FFNNs in GPS and INS integration for vehicular navigation[C]∥2011 International Conference on Recent Trends in Information Technology (ICRTIT), Chennai, India, 2011: 223-228.
13 Malleswaran M, Vaidehi V, Saravanaselvan A, et al. Performance analysis of various artificial intelligent neural networks for GPS/INS integration[J]. Applied Artificial Intelligence, 2013, 27(5): 367-407.
14 Dai H, Bian H, Wang R, et al. An INS/GNSS integrated navigation in GNSS denied environment using recurrent neural network[J]. Defence Technology, 2020, 16(2): 334-340.
15 Fang W, Jiang J, Lu S, et al. A LSTM algorithm estimating pseudo measurements for aiding INS during GNSS signal outages[J]. Remote Sensing, 2020, 12(2): No.256.
16 Onyekpe U, Palade V, Kanarachos S, et al. Learning uncertainties in wheel odometry for vehicular localisation in GNSS deprived environments[C]∥2020 19th IEEE International Conference on Machine Learning and Applications (ICMLA), Miami, FL, USA, 2020: 741-746.
17 Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need[J/OL]. [2022-09-28].
18 Onyekpe U, Kanarachos S, Palade V, et al. Vehicular localisation at high and low estimation rates during gnss outages: a deep learning approach[J]. Deep Learning Applications, 2021,2: 237-256.
19 Onyekpe U, Palade V, Kanarachos S, et al. IO-VNBD: inertial and odometry benchmark dataset for ground vehicle positioning[J]. Data in Brief, 2021, 35: No.106885.
20 Onyekpe U, Palade V, Kanarachos S. Learning to localise automated vehicles in challenging environments using Inertial Navigation Systems (INS)[J]. Applied Sciences, 2021, 11(3): No.1270.
[1] 王登峰,陈宏利,那景新,陈鑫. 单双搭接接头经高温老化后的失效对比[J]. 吉林大学学报(工学版), 2023, 53(2): 346-354.
[2] 张佩,王志伟,杜常清,颜伏伍,卢炽华. 车用质子交换膜燃料电池空气系统过氧比控制方法[J]. 吉林大学学报(工学版), 2022, 52(9): 1996-2003.
[3] 王克勇,鲍大同,周苏. 基于数据驱动的车用燃料电池故障在线自适应诊断算法[J]. 吉林大学学报(工学版), 2022, 52(9): 2107-2118.
[4] 曹起铭,闵海涛,孙维毅,于远彬,蒋俊宇. 质子交换膜燃料电池低温启动水热平衡特性[J]. 吉林大学学报(工学版), 2022, 52(9): 2139-2146.
[5] 隗海林,王泽钊,张家祯,刘洋. 基于Avl-Cruise的燃料电池汽车传动比及能量管理策略[J]. 吉林大学学报(工学版), 2022, 52(9): 2119-2129.
[6] 刘岩,丁天威,王宇鹏,都京,赵洪辉. 基于自适应控制的燃料电池发动机热管理策略[J]. 吉林大学学报(工学版), 2022, 52(9): 2168-2174.
[7] 高金武,贾志桓,王向阳,邢浩. 基于PSO-LSTM的质子交换膜燃料电池退化趋势预测[J]. 吉林大学学报(工学版), 2022, 52(9): 2192-2202.
[8] 李丞,景浩,胡广地,刘晓东,冯彪. 适用于质子交换膜燃料电池系统的高阶滑模观测器[J]. 吉林大学学报(工学版), 2022, 52(9): 2203-2212.
[9] 池训逞,侯中军,魏伟,夏增刚,庄琳琳,郭荣. 基于模型的质子交换膜燃料电池系统阳极气体浓度估计技术综述[J]. 吉林大学学报(工学版), 2022, 52(9): 1957-1970.
[10] 裴尧旺,陈凤祥,胡哲,翟双,裴冯来,张卫东,焦杰然. 基于自适应LQR控制的质子交换膜燃料电池热管理系统温度控制[J]. 吉林大学学报(工学版), 2022, 52(9): 2014-2024.
[11] 胡广地,景浩,李丞,冯彪,刘晓东. 基于高阶燃料电池模型的多目标滑模控制[J]. 吉林大学学报(工学版), 2022, 52(9): 2182-2191.
[12] 陈凤祥,伍琪,李元松,莫天德,李煜,黄李平,苏建红,张卫东. 2.5吨燃料电池混合动力叉车匹配、仿真及优化[J]. 吉林大学学报(工学版), 2022, 52(9): 2044-2054.
[13] 武小花,余忠伟,朱张玲,高新梅. 燃料电池公交车模糊能量管理策略[J]. 吉林大学学报(工学版), 2022, 52(9): 2077-2084.
[14] 高青,王浩东,刘玉彬,金石,陈宇. 动力电池应急冷却喷射模式实验分析[J]. 吉林大学学报(工学版), 2022, 52(8): 1733-1740.
[15] 王奎洋,何仁. 基于支持向量机的制动意图识别方法[J]. 吉林大学学报(工学版), 2022, 52(8): 1770-1776.
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
No Suggested Reading articles found!