Research map / Adversarial robustness and security / Certified and empirical robustness
On Adversarial Robustness of Trajectory Prediction for Autonomous Vehicles
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Apply physically constrained white-box adversarial perturbations to observed trajectories | GRIP, FQA, and Trajectron++ · Apolloscape, NGSIM, and nuScenes | ADE/FDE: ADE/FDE is increased by 167%/150% | Helped |
| Add map information as input | Trajectron++ · nuScenes | ADE: reduced by 22% and 31% | Hurt |
| Extend the adversarial perturbation across a 3-second prediction period instead of a single frame (ablation) | White-box attack on three models · Apolloscape | ADE/FDE: 142%/127% | Hurt |
| Reduce the maximum trajectory-location deviation bound to 0.2-meter (ablation) | White-box attack · Apolloscape | ADE/FDE: 86%/80% | Hurt |
| Drop 50% or 100% of objects other than the target vehicle (ablation) | White-box attack on three models · Apolloscape | prediction error | No clear effect |
| Use data augmentation with constrained random trajectory perturbations during training | Trajectron++ · Apolloscape | prediction error under attack: reduced by 24% | Hurt |
| Apply trajectory smoothing during training | FQA · Apolloscape | prediction error under attack | Hurt |
| Combine data augmentation and trajectory smoothing during training | Three prediction models · Apolloscape | prediction error under attack: reduced by 26% on average | Hurt |
| Combine data augmentation and trajectory smoothing during training | Three prediction models · Apolloscape normal cases | prediction error: increased by 11% | Helped |
| Smooth every trajectory at test time | Three prediction models · Apolloscape attacked cases | prediction error: dropped by 13% | Hurt |
| Smooth every trajectory at test time | Three prediction models · Apolloscape normal cases | prediction error: raised significantly by 28% | Helped |
| Detect adversarial trajectories and smooth only detected cases at test time | Three prediction models · Apolloscape | prediction error under attack: reduced by 12% | Hurt |
Closest papers by meaning
- Enduring, Efficient and Robust Trajectory Prediction Attack in Autonomous Driving via Optimization-Driven Multi-Frame Perturbation Framework (CVPR 2025 · similarity 0.91)
- Adversarial Backdoor Attack by Naturalistic Data Poisoning on Trajectory Prediction in Autonomous Driving (CVPR 2024 · similarity 0.90)
- Certified Human Trajectory Prediction (CVPR 2025 · similarity 0.89)
- Certified Adversarial Robustness via Randomized $\alpha$-Smoothing for Regression Models (NeurIPS 2024 · similarity 0.85)
- CaDeT: a Causal Disentanglement Approach for Robust Trajectory Prediction in Autonomous Driving (CVPR 2024 · similarity 0.85)
- World Models as Adversaries: Multi-Agent Self-Play Fine-Tuning for Robust Motion Planning (NeurIPS 2026 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.