Research map / Adversarial robustness and security / Attacks on text and graphs
SlowLiDAR: Increasing the Latency of LiDAR-Based Detection Using Adversarial Examples
Read the paperCheck 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 |
|---|---|---|---|
| Use latency-targeted adversarial point perturbations instead of integrity-targeted PGD, CW, or I-FGM | LiDAR detection models · KITTI | ROI-P: PIXOR: 35.21 vs 0.05; YOLO3D: 499.00 vs 0.00; Complex-YOLO: 539.00 vs 1.00; VoxelNet: 48.29 vs 0.25; PointPillar: 165.92 | Helped |
| Add adversarial points to increase proposals before NMS | LiDAR detection models · KITTI | ROI-P: PIXOR: 45.15 vs 0.55; YOLO3D: 479.00 vs 2.50; Complex-YOLO: 552.00 vs 5.00; VoxelNet: 33.38 vs 0.83; PointPillar: 156.16 | Helped |
| Use SlowLiDAR point perturbation to increase end-to-end latency | LiDAR detection models · KITTI, 3090 Ti | ROI-TT: PIXOR 85.08; YOLO3D 327.33; Complex-YOLO 136.76; VoxelNet 2.88; PointPillar 3.12; Second 4.22 | Helped |
| Use SlowLiDAR point addition to increase end-to-end latency | LiDAR detection models · KITTI, 3090 Ti | ROI-TT: PIXOR 48.22; YOLO3D 94.73; Complex-YOLO 36.41; VoxelNet 1.92; PointPillar 2.30; Second 2.68 | Helped |
| Use ensemble min-max training rather than training on one source model (ablation) | SlowLiDAR transfer attacks · PIXOR | ROI-P: 12.994 | Helped |
| Increase the number of effective probing steps for point initialization (ablation) | SlowLiDAR point addition · KITTI | ROI-P | Helped |
| Increase effective probing steps (ablation) | SlowLiDAR point addition · KITTI | processing time | Hurt |
| Increase the proposal-limit or confidence-score threshold (ablation) | NMS defenses against SlowLiDAR · KITTI | ROI-NMS: 47.28; 88.9 | Mixed |
| Add Gaussian noise to adversarial point clouds (ablation) | SlowLiDAR attacks · KITTI, noise standard deviation 0.01m | ROD-Proposal: less than 3%; less than 13% | No clear effect |
| Constrain perturbation updates to a discrete laser-ray resolution (ablation) | SlowLiDAR physical-realizability variant · KITTI | ROI-P: PointPillar 90.016; VoxelNet 23.209; PIXOR 12.979 | Helped |
| Transfer a sample crafted on a different model instead of optimizing directly for the target (ablation) | SlowLiDAR · KITTI, cross-model transfer | ROI-P: PIXOR-to-VoxelNet 1.722 vs 48.286; VoxelNet-to-PIXOR 1.115 vs 35.213 | Hurt |
| Raise the proposal limit or confidence threshold as a defense (ablation) | LiDAR detection models · KITTI | benign detection performance | Hurt |
Closest papers by meaning
- SlowTrack: Increasing the Latency of Camera-Based Perception in Autonomous Driving Using Adversarial Examples (AAAI 2024 · similarity 0.89)
- A New Adversarial Perspective for LiDAR-based 3D Object Detection (AAAI 2025 · similarity 0.89)
- Enduring, Efficient and Robust Trajectory Prediction Attack in Autonomous Driving via Optimization-Driven Multi-Frame Perturbation Framework (CVPR 2025 · similarity 0.88)
- Neural Reconstruction of LiDAR Point Clouds under Jamming Attacks via Full-Waveform Representation and Simultaneous Laser Sensing (NeurIPS 2026 · similarity 0.86)
- An Empirical Study of the Generalization Ability of Lidar 3D Object Detectors to Unseen Domains (CVPR 2024 · similarity 0.86)
- LiON: Learning Point-Wise Abstaining Penalty for LiDAR Outlier DetectioN Using Diverse Synthetic Data (AAAI 2025 · 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.