Automotive Engineering ›› 2024, Vol. 46 ›› Issue (3): 407-417.doi: 10.19562/j.chinasae.qcgc.2024.03.004
Haifeng Sang,Zishan Zhao(),Jinyu Wang,Wangxing Chen
Received:
2023-08-03
Revised:
2023-09-19
Online:
2024-03-25
Published:
2024-03-18
Contact:
Zishan Zhao
E-mail:zhao_zishan@smail.sut.edu.cn
Haifeng Sang,Zishan Zhao,Jinyu Wang,Wangxing Chen. Research on Adversarial Attacks and Robustness in Vehicle Trajectory Prediction[J].Automotive Engineering, 2024, 46(3): 407-417.
"
模型 | 数据集 | 阈值 | |||||
---|---|---|---|---|---|---|---|
Ade | Fde | Front | Left | Rear | Right | ||
GRIP++ | Apolloscape | 0.873 0 | 1.338 6 | 0.795 8 | 0.408 9 | 0.950 8 | 0.383 4 |
GRIP++ | NGSIM | 0.168 0 | 0.370 7 | 0.298 9 | 0.109 2 | 0.205 3 | 0.101 5 |
GRIP++ | nuScenes | 0.342 8 | 0.728 9 | 0.447 5 | 0.221 3 | 0.468 4 | 0.186 7 |
FQA | Apolloscape | 1.172 8 | 1.848 9 | 1.279 1 | 0.392 2 | 1.287 7 | 0.436 4 |
FQA | NGSIM | 0.168 0 | 0.370 7 | 0.289 3 | 0.154 2 | 0.280 9 | 0.141 3 |
FQA | nuScenes | 0.409 6 | 0.711 1 | 0.368 3 | 0.195 6 | 0.396 7 | 0.172 8 |
Trajectron++ | Apolloscape | 1.727 6 | 1.120 6 | 1.227 8 | 0.390 8 | 1.148 8 | 0.453 6 |
Trajectron++ | NGSIM | 0.284 8 | 0.540 5 | 0.178 2 | 0.151 5 | 0.323 7 | 0.214 7 |
Trajectron++ | nuScenes | 0.410 4 | 0.583 5 | 0.614 3 | 0.351 8 | 0.657 4 | 0.285 6 |
"
Model | Dataset | ADE | FDE | Left | ||||||
---|---|---|---|---|---|---|---|---|---|---|
Clean | Search | PTFA | Clean | Search | PTFA | Clean | Search | PTFA | ||
GRIP++ | Apolloscape | -1.978 6 | -7.082 7 | -7.006 9 | -3.183 5 | -11.055 8 | -10.875 1 | 0.012 9 | -2.536 1 | |
GRIP++ | NGSIM | -3.293 6 | -3.847 7 | -6.827 5 | -7.483 7 | 0.176 8 | -0.433 8 | |||
GRIP++ | nuScenes | -5.462 3 | -8.039 3 | -10.394 2 | -15.153 2 | -0.232 1 | -1.409 5 | |||
FQA | Apolloscape | -2.378 9 | -9.074 3 | -3.831 5 | -14.166 5 | -0.048 3 | -2.457 2 | |||
FQA | NGSIM | -5.684 3 | -6.625 6 | -9.863 3 | -11.865 1 | -0.179 1 | -1.013 8 | |||
FQA | nuScenes | -5.283 2 | -7.655 9 | -10.039 8 | -14.140 4 | -0.229 8 | -1.181 4 | |||
Trajectron++ | Apolloscape | -3.792 8 | -9.952 9 | -6.000 0 | -15.396 2 | -0.152 3 | -2.602 8 | -2.586 7 | ||
Trajectron++ | NGSIM | -23.285 0 | -30.380 0 | -41.629 0 | -53.872 0 | 0.466 7 | -1.260 5 | |||
Trajectron++ | nuScenes | -8.755 0 | -13.211 0 | -17.054 0 | -25.057 0 | 0.339 4 | -4.256 1 | |||
Model | Dataset | Right | Front | Rear | ||||||
Clean | Search | PTFA | Clean | Search | PTFA | Clean | Search | PTFA | ||
GRIP++ | Apolloscape | -0.012 9 | -2.444 2 | -2.442 8 | 0.015 3 | -4.752 9 | -0.015 3 | -5.310 9 | ||
GRIP++ | NGSIM | -0.176 8 | -0.664 1 | 5.301 8 | 0.018 6 | -5.301 8 | -2.668 8 | |||
GRIP++ | nuScenes | 0.232 1 | -0.991 0 | 1.044 8 | -2.036 0 | -1.044 8 | -3.904 7 | |||
FQA | Apolloscape | 0.048 3 | -2.752 4 | -2.735 0 | 0.389 1 | -6.705 7 | -6.696 0 | -3.831 5 | -7.531 2 | -7.520 6 |
FQA | NGSIM | 0.179 1 | -0.605 4 | 1.368 5 | -0.113 9 | -1.368 5 | -2.793 3 | |||
FQA | nuScenes | 0.229 8 | -0.736 0 | 0.815 0 | -1.797 4 | -0.815 0 | -3.348 0 | |||
Trajectron++ | Apolloscape | 0.152 3 | -2.598 9 | -2.594 5 | 0.351 2 | -6.436 3 | -0.351 2 | -6.844 0 | -6.829 4 | |
Trajectron++ | NGSIM | -0.466 7 | -0.734 3 | -23.463 | -27.031 0 | 19.463 0 | 17.565 0 | |||
Trajectron++ | nuScenes | -0.339 4 | -0.465 1 | 1.890 7 | -2.244 5 | -1.890 7 | -6.014 7 |
"
模型 | 数据集 | ADE | FDE | Left | ||||||
---|---|---|---|---|---|---|---|---|---|---|
Search | DLRA | diff/% | Search | DLRA | diff/% | Search | DLRA | diff/% | ||
GRIP++ | Apolloscape | -6.75 | -7.00 | -10.59 | -10.94 | -2.35 | -2.36 | |||
NGSIM | -3.35 | -4.05 | -6.54 | -7.24 | -0.43 | -0.79 | ||||
nuScenes | -7.99 | -8.11 | -14.31 | -14.45 | -1.24 | -1.40 | ||||
FQA | Apolloscape | -8.76 | -9.45 | -13.84 | -14.72 | -2.40 | -2.41 | |||
NGSIM | -4.52 | -5.15 | -8.37 | -9.07 | -1.27 | -2.31 | ||||
nuScenes | -6.68 | -6.75 | -11.51 | -11.52 | -1.06 | -1.22 | ||||
Trajectron++ | Apolloscape | -9.36 | -9.62 | -14.61 | -14.99 | -2.32 | -2.32 | |||
NGSIM | -29.91 | -33.71 | -53.72 | -56.74 | -1.22 | -1.69 | ||||
nuScenes | -13.21 | -13.36 | -24.94 | -25.05 | -4.09 | -4.12 | ||||
模型 | 数据集 | Right | Front | Rear | ||||||
Search | DLRA | diff/% | Search | DLRA | diff/% | Search | DLRA | diff/% | ||
GRIP++ | Apolloscape | -2.25 | -2.25 | -4.71 | -4.79 | -5.31 | -5.43 | |||
NGSIM | -0.65 | -1.05 | 0.09 | -0.59 | -2.69 | -3.03 | ||||
nuScenes | -1.27 | -1.44 | -1.77 | -1.91 | -3.87 | -3.88 | ||||
FQA | Apolloscape | -2.66 | -2.65 | -0.22 | -6.80 | -6.81 | -7.70 | -7.73 | ||
NGSIM | -0.11 | -1.47 | -1.52 | -1.98 | -1.21 | -1.69 | ||||
nuScenes | -1.25 | -1.29 | -1.70 | -1.85 | -3.62 | -3.98 | ||||
Trajectron++ | Apolloscape | -2.49 | -2.52 | -6.97 | -6.98 | -6.16 | -6.16 | 0.00 | ||
NGSIM | -0.79 | -1.12 | -27.70 | -28.17 | 19.23 | 16.39 | ||||
nuScenes | -1.98 | -1.98 | 0.00 | -0.81 | -0.88 | -6.01 | -6.22 |
"
模型 | 数据集 | Search | PTFA | DLRA | |||
---|---|---|---|---|---|---|---|
时间/s | 不易被攻击的样本个数 | 时间/s | 不易被攻击的样本个数 | 时间/s | 不易被攻击的样本个数 | ||
GRIP++ | Apolloscape | 155.53 | 5 | 26.43 | 1 | 145.32 | 1 |
GRIP++ | NGSIM | 741.87 | 12 | 194.63 | 1 | 604.52 | 1 |
GRIP++ | nuScenes | 212.48 | 12 | 71.36 | 3 | 197.79 | 3 |
FQA | Apolloscape | 202.14 | 7 | 38.34 | 2 | 172.61 | 2 |
FQA | NGSIM | 740.79 | 16 | 213.40 | 4 | 645.03 | 4 |
FQA | nuScenes | 335.07 | 35 | 180.58 | 7 | 313.05 | 7 |
Trajectron++ | Apolloscape | 6 366.25 | 15 | 1 383.21 | 10 | 4 727.87 | 10 |
Trajectron++ | NGSIM | 85 694.35 | 3 | 10 964.11 | 0 | 69 437.48 | 0 |
Trajectron++ | nuScenes | 25 129.41 | 34 | 11 022.48 | 13 | 15 175.44 | 13 |
"
防御策略 | 对抗样本 | 预测误差(误差变化) | |||||
---|---|---|---|---|---|---|---|
ADE | FDE | Left | Right | Front | Rear | ||
数据增强 | 干净数据 | -2.296(-0.083) | -3.704(-0.127) | 0.0576(-0.106) | -0.0576(-0.106) | 0.182368(-0.207) | -0.182(-0.207) |
随机攻击 | -5.130(-0.065) | -7.970(-0.107) | -1.226(-0.101) | -2.864(-0.193) | |||
Search | -8.901(-0.173) | -14.015(-0.151) | -2.378(-0.079) | -7.355(-0.176) | |||
DLRA | -9.476(-0.142) | -14.690(-0.188) | -2.364(-0.077) | -7.338(-0.182) | |||
卷积平滑 | 干净数据 | -2.104(-0.275) | -3.394(-0.437) | -0.021(-0.027) | 0.021(-0.027) | 0.295(-0.094) | -0.295(-0.094) |
随机攻击 | -3.701(-1.494) | -5.752(-2.325) | -0.861(-0.466) | -0.880(-0.340) | -1.072(-0.789) | -1.858(-1.199) | |
Search | -6.050(-2.851) | -9.522(-4.493) | -2.155(-0.222) | -2.277(-0.634) | -3.994(-2.951) | -4.614(-2.740) | |
DLRA | -6.379(-3.097) | -9.841(-4.849) | -2.140(-0.224) | -2.264(-0.621) | -3.977(-2.965) | -4.587(-2.751) |
1 | SONG H, DING W, CHEN Y, et al. Pip: planning-informed trajectory prediction for autonomous driving[C].Computer Vision-ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XXI 16. Springer International Publishing, 2020: 598-614. |
2 | 桑海峰,陈旺兴,王海峰,等.基于多模式时空交互的行人轨迹预测模型[J].电子学报,2022,50(11):2806-2812. |
SANG H F, CHEN W X, WANG H F, et al. Pedestrian trajectory prediction model based on multi-modal spatio-temporal interaction[J]. Journal of Electronics, 2022, 50(11): 2806-2812. | |
3 | LI X, YING X, MOOI C C. Grip++: enhanced graph-based interaction aware trajectory prediction for autonomous driving[EB/OL]. [2019-07-17].https://arxiv.org/pdf/1907.07792.pdf. |
4 | SANG H, CHEN W, WANG J, et al. RDGCN: reasonably dense graph convolution network for pedestrian trajectory prediction[J]. Measurement, 2023, 213: 112675. |
5 | BAHARI M, SAADATNEJAD S, RAHIMI A, et al. Vehicle trajectory prediction works, but not everywhere[C].Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022: 17123-17133. |
6 | SAADATNEJAD S, BAHARI M, KHORSANDI P, et al. Are socially-aware trajectory prediction models really socially-aware?[J]. Transportation Research Part C: Emerging Technologies, 2022, 141: 103705. |
7 | CAO Y, XU D, WENG X, et al. Robust trajectory prediction against adversarial attacks[C].Conference on Robot Learning. PMLR, 2023: 128-137. |
8 | CAO Y, XIAO C, ANANDKUMAR A, et al. AdvDO: realistic adversarial attacks for trajectory prediction[C] .European Conference on Computer Vision. Springer, Cham, 2022: 36-52. |
9 | ETTINGER S, CHENG S, CAINE B, et al. Large scale interactive motion forecasting for autonomous driving: The waymo open motion dataset[C].Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021: 9710-9719. |
10 | MI L, ZHAO H, NASH C, et al. HDMapGen: a hierarchical graph generative model of high definition maps[C].Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021: 4227-4236. |
11 | GUPTA A, JOHNSON J, FEI-FEI L, et al. Social GAN: socially acceptable trajectories with generative adversarial networks[C].Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 2255-2264. |
12 | ZHANG Q, HU S, SUN J, et al. On adversarial robustness of trajectory prediction for autonomous vehicles[C].Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022: 15159-15168. |
13 | MOOSAVI-DEZFOOLI S M, FAWZI A, FROSSARD P. DeepFool: a simple and accurate method to fool deep neural networks[C].Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 2574-2582. |
14 | LI Y, WEN C, JUEFEI-XU F, et al. Fooling lidar perception via adversarial trajectory perturbation[C].Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021: 7898-7907. |
15 | WANG J, PUN A, TU J, et al. AdvSim: generating safety-critical scenarios for self-driving vehicles[C].Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021: 9909-9918. |
16 | HUANG X, CHENG X, GENG Q, et al. The apolloscape dataset for autonomous driving[C].Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, 2018: 954-960. |
17 | CAESAR H, BANKITI V, LANG A H, et al. nuScenes: a multimodal dataset for autonomous driving[C].Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 11621-11631. |
18 | COIFMAN B, LI L. A critical evaluation of the Next Generation Simulation (NGSIM) vehicle trajectory dataset[J]. Transportation Research Part B: Methodological, 2017, 105: 362-377. |
19 | KAMRA N, ZHU H, TRIVEDI D K, et al. Multi-agent trajectory prediction with fuzzy query attention[J]. Advances in Neural Information Processing Systems, 2020, 33: 22530-22541. |
20 | SALZMANN T, IVANOVIC B, CHAKRAVARTY P, et al. Trajectron++: dynamically-feasible trajectory forecasting with heterogeneous data[C].Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XVIII 16. Springer International Publishing, 2020: 683-700. |
[1] | Yiwei Zhou,Mo Xia,Bing Zhu. Multimodal Vehicle Trajectory Prediction Methods Considering Multiple Traffic Participants in Urban Road Scenarios [J]. Automotive Engineering, 2024, 46(3): 396-406. |
[2] | Xinzheng Wu,Xingyu Xing,Lihao Liu,Yong Shen,Junyi Chen. Testing and Analysis of the Robustness of Decision-Making and Planning Systems Based on Fault Injection [J]. Automotive Engineering, 2023, 45(8): 1428-1437. |
[3] | Zhangpeng Gong,Guoye Wang,Shi Yu. The Algorithm of Multi⁃Category Object Recognition in Road Scene Based on Voxel Network [J]. Automotive Engineering, 2021, 43(4): 469-477. |
[4] | Haisheng Song,Na Yang,Changhai Yang,Zhiyong Chen,Wenku Shi. Multi⁃object Adaptive Optimization of Spot Welds with Consideration of Robustness [J]. Automotive Engineering, 2021, 43(3): 442-449. |
[5] | Jin Bowen, Qiao Huimin, Pan Tianhong, Chen Shan. Lithium Battery SOC Estimation Based on Internal Resistance Power Consumption [J]. Automotive Engineering, 2020, 42(8): 1008-1015. |
[6] | Zhang Haiyang, Lü Xiaojiang, Zhou Dayong, Xia Liang, Gu Xianguang. Optimization Design of Vehicle Structural Crashworthiness with Consideration of Robustness [J]. Automotive Engineering, 2020, 42(2): 222-227. |
[7] | Song Haisheng, Chen Zhiyong, Yang Na, Yang Changhai, Shi Wenku. Research on the Analysis Method of Welding Spot Layout Robustness Based on Sensitivity Ranking [J]. Automotive Engineering, 2020, 42(12): 1751-1757. |
[8] | Xie Feng, Lou Jingtao, Zhao Kai & Qi Yao. A Research on Vehicle Trajectory Prediction Method Based onBehavior Recognition and Curvature Constraints [J]. Automotive Engineering, 2019, 41(9): 1036-1042. |
[9] | Zhao Xi, Chen Shuai, Ying Liang, Hou Wenbin, Hu Ping. Crashworthiness Optimization of Automotive Thin-walled Structure with Functionally Graded Strength#br# [J]. , 2018, 40(5): 508-514. |
|