Explainability-Aware Frustum Attack: Exposing Structural Vulnerabilities in LiDAR-Based 3D Object Detectors
Chengzeng You, Binbin Xu, Soteris Demetriou · 2026
Read Explainability-Aware Frustum Attack: Exposing Structural Vulnerabilities in LiDAR-Based 3D Object Detectors by Chengzeng You, Binbin Xu, Soteris Demetriou (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-29963.txt.
Description
You, Xu, and Demetriou probe how LiDAR-based 3D object detectors, the perception systems of self-driving cars, can be fooled efficiently. Earlier attacks either studied isolated objects or spoofed whole scenes for physical realism, without asking which spatial evidence the detector actually relies on.
Their Saliency-LiDAR method aggregates gradient-based attributions across scenes into universal saliency maps showing the regions detectors depend on. Guided by these maps, the Explainability-aware Frustum Attack perturbs only the most influential frustums, cone-shaped sensor regions, instead of whole objects. On KITTI and nuScenes with detectors such as PointPillars and SECOND, detection recall falls by more than 15 points while using 25 to 50 percent fewer perturbed frustums than the leading baseline.
Catalog record
| CREATOR | Chengzeng You, Binbin Xu, Soteris Demetriou |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 45 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.29963