Laser attack blinds autonomous vehicles, deleting pedestrians and confusing cars

lidarautonomous-vehiclesadversarial-attacksensor-spoofingphysical-security

Abstraction: Timed laser attacks spoof lidar sensors to erase pedestrians from AV perception

Key points:

Researchers from University of Florida, University of Michigan, and University of Electro-Communications (Japan) demonstrated the first attack that causes lidar sensors to delete obstacle data rather than just inject false objects
Attacker ~15 feet away fires precisely timed laser pulses that mimic lidar reflections, causing the sensor to discard genuine returns from real obstacles as duplicates
In autonomous vehicle decision-making simulations, the blind spot caused cars to continue accelerating toward hidden pedestrians instead of stopping
Attack works on moving vehicles using basic camera tracking equipment; timing information needed is publicly available from lidar manufacturers
Researchers also demonstrated following a slow-moving vehicle in real-world experiments using the attack
Proposed mitigations: teach sensor software to detect spoofed reflection signatures; to be presented at 2023 USENIX Security Symposium (arXiv:2210.09482)

Connections: Autonomous Vehicles · Adversarial Examples · AI Safety