Reproduce “FlyTrap: Physical Distance-Pulling Attack Towards Camera-based Autonomous Target Tracking Systems” 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
Shaoyuan Xie, Mohamad Habib Fakih, Junchi Lu, Fayzah Alshammari, Ningfei Wang, Takami Sato, Halima Bouzidi, Mohammad Abdullah Al Faruque and 1 more
A physical attack framework, FlyTrap, enables 'distance-pulling attacks' (DPA) on camera-based autonomous target tracking systems, manipulating them to dangerously reduce tracking distance. Researchers at the University of California, Irvine, demonstrated that an adversarial umbrella could facilitate drone capture, range-limited sensor attacks, or physical crashes against custom and commercial drones, effectively bypassing current spatial-temporal consistency defenses.
Key insights
- Manipulating tracking distance in ATT systems, rather than just disrupting tracking, enables severe physical consequences such as drone capture, collision, or expanded opportunities for range-limited sensor attacks.
- The use of an inconspicuous, physically realizable adversarial vector, like an umbrella, combined with closed-loop simulation and consistency constraints, allows for effective and stealthy attacks in real-world outdoor environments.
- Defenses relying on spatial-temporal consistency checks can be bypassed by attacks that explicitly co-optimize both spatial features (e.g., human-like bounding box, pose) and gradual temporal changes in perceived target size.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.