Reproduce “DisorientLiDAR: Physical Attacks on LiDAR-based Localization” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Yizhen Lao, Yu Zhang, Ziting Wang, Chengbo Wang, Yifei Xue, Wanpeng Shao

DisorientLiDAR attacks LiDAR-based localization by using a registration model to identify useful scene regions, then covering them with near-infrared-absorbing material to suppress returns. On KITTI, reported D3Feat Top-K results show registration recall falling from 0.729 to 0.569, with translation and rotation errors rising. Physical tests disrupted selected scan-registration cases, but did not demonstrate a live vehicle being driven off course.

Key insights

  • LiDAR-based localization models, particularly deep learning-based ones, are highly susceptible to targeted removal of a minimal number of critical geometric 'key regions,' leading to significant and persistent localization errors.
  • Physical adversarial attacks can be executed remotely and stealthily using readily available NIR-absorbing materials, posing a practical threat to autonomous vehicles.
  • The attack is effective against various LiDAR hardware configurations and both deep-learning-based and traditional localization algorithms (like NDT), demonstrating a broad and generalized vulnerability.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.