On Adversarial Robustness of Trajectory Prediction for Autonomous Vehicles

Zhang, Qingzhao; Hu, Shengtuo; Sun, Jiachen; Chen, Qi Alfred; Mao, Z. Morley

doi:10.1109/cvpr52688.2022.01473

Cited by 54 publications

(25 citation statements)

References 68 publications

(134 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Although AdvSim enforces the dynamic feasibility of the synthesized trajectories, it uses black-box optimization which is slow and unreliable. Our work is most similar to a very recent work [18]. However, as we will show empirically, [18] fails to generate dynamically feasible adversarial trajectories.…”

Section: Related Worksupporting

confidence: 55%

“…Our work is most similar to a very recent work [18]. However, as we will show empirically, [18] fails to generate dynamically feasible adversarial trajectories. This is because its threat model simply uses dataset statistics (e.g.…”

Section: Related Worksupporting

confidence: 55%

“…There is a large body of literature on improving their adversarial robustness [31][32][33][34][35][36][37][38][39][40][41][42][43][44]. In the AV context, many works examine on the adversarial robustness of the perception task [45], while analyzing the adversarial robustness of trajectory forecaster [18] is rarely explored. In this work, we focus on studying the adversarial robustness in the trajectory prediction task by considering its unique properties including motion and social interaction.…”

Section: Related Workmentioning

confidence: 99%

“…Baselines. We select the search-based attack proposed by Zhang et al [18] as the baseline, named search. As we mentioned earlier in § 2, the original method made two mistakes: (1) incorrect estimated bound values for dynamic parameters and (2) incorrect choices of bounded dynamic parameters for generating realistic adversarial trajectories.…”

Section: Experimental Settingmentioning

confidence: 99%

See 3 more Smart Citations

AdvDO: Realistic Adversarial Attacks for Trajectory Prediction

Cao

Xiao

Anandkumar

et al. 2022

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Trajectory prediction is essential for autonomous vehicles (AVs) to plan correct and safe driving behaviors. While many prior works aim to achieve higher prediction accuracy, few study the adversarial robustness of their methods. To bridge this gap, we propose to study the adversarial robustness of data-driven trajectory prediction systems. We devise an optimization-based adversarial attack framework that leverages a carefully-designed differentiable dynamic model to generate realistic adversarial trajectories. Empirically, we benchmark the adversarial robustness of state-of-the-art prediction models and show that our attack increases the prediction error for both general metrics and planningaware metrics by more than 50% and 37%. We also show that our attack can lead an AV to drive off road or collide into other vehicles in simulation. Finally, we demonstrate how to mitigate the adversarial attacks using an adversarial training scheme 1 .

show abstract

Section: Related Worksupporting

confidence: 55%

Section: Related Worksupporting

confidence: 55%

Section: Related Workmentioning

confidence: 99%

Section: Experimental Settingmentioning

confidence: 99%

See 2 more Smart Citations

AdvDO: Realistic Adversarial Attacks for Trajectory Prediction

Cao

Xiao

Anandkumar

et al. 2022

Lecture Notes in Computer Science

View full text Add to dashboard Cite

show abstract

“…3D data is used in many different fields, including autonomous driving, robotics, remote sensing, and more [5,12,14,17,47]. Point cloud has a very uniform structure, which avoids the irregularity and complexity of composition.…”

Section: Introductionmentioning

confidence: 99%

ProxyFormer: Proxy Alignment Assisted Point Cloud Completion with Missing Part Sensitive Transformer

Li¹,

Gao²,

Tan³

et al. 2023

Preprint

View full text Add to dashboard Cite

Problems such as equipment defects or limited viewpoints will lead the captured point clouds to be incomplete. Therefore, recovering the complete point clouds from the partial ones plays an vital role in many practical tasks, and one of the keys lies in the prediction of the missing part. In this paper, we propose a novel point cloud completion approach namely ProxyFormer that divides point clouds into existing (input) and missing (to be predicted) parts and each part communicates information through its proxies. Specifically, we fuse information into point proxy via feature and position extractor, and generate features for missing point proxies from the features of existing point proxies. Then, in order to better perceive the position of missing points, we design a missing part sensitive transformer, which converts random normal distribution into reasonable position information, and uses proxy alignment to refine the missing proxies. It makes the predicted point proxies more sensitive to the features and positions of the missing part, and thus make these proxies more suitable for subsequent coarse-to-fine processes. Experimental results show that our method outperforms state-of-the-art completion networks on several benchmark datasets and has the fastest inference speed. Code is available at https://github. com/I2-Multimedia-Lab/ProxyFormer.

show abstract