The laser attack blinds autonomous vehicles, removes pedestrians and confuses cars

Self-driving cars, like human drivers before them, need to see what’s around them to avoid obstacles and drive safely.

The most advanced autonomous vehicles typically use lidar, a rotating radar-type device that acts as the car’s eyes. Lidar provides continuous information about the distance to objects so that the car can decide which actions are safe to take.

But these eyes can be fooled.

New research reveals that expertly timed lasers that shine on an approaching lidar system can create a blind spot in front of a vehicle large enough to hide moving pedestrians and other obstacles. The deleted information makes cars think the road is safe to continue on, putting any potential attack in the blind spot at risk.

This is the first time that lidar sensors have been tricked into removing information about obstacles.

The vulnerability was discovered by researchers from the University of Florida, the University of Michigan and Japan’s University of Electronic Communications. Scientists also provide updates that can remove this weakness to protect people from malicious attacks.

The results will be presented at the 2023 USENIX Security Symposium and are currently being published arXiv.

Lidar works by sending out laser light and capturing the reflections to calculate distances, just as a bat echolocator uses sound echoes. The attack creates false reflections to confuse the sensor.

“We mimic lidar reflections with our laser so that the sensor reduces other reflections that come from real obstacles,” said Sara Rampazzi, a UF professor of computing and information technology who led the research. “The Lidar still receives genuine data from the obstacle, but the data is automatically discarded because the sensor detects our fake reflections.”

The researchers demonstrated the attack on moving vehicles and robots by placing the attacker about 15 meters from the edge of the road. But in theory it could be achieved even further with updated devices. The technology required is all fairly simple, but the laser needs to be perfectly timed to the lidar sensor and moving vehicles need to be carefully tracked to keep the laser pointing in the right direction.

“It’s primarily about synchronizing the laser with the lidar. The information you need is usually publicly available from the manufacturer,” said S. Hrushikesh Bhupathiraj, a UF doctoral student in Rampazzi’s lab and one of the study’s lead authors. .

Using this technique, researchers were able to remove data from static obstacles and moving pedestrians. They also demonstrated in real experiments that an attack could track a slow-moving vehicle using basic camera trackers. In simulations of autonomous vehicle decision-making, this removal of data caused the car to continue accelerating toward a pedestrian it no longer saw, instead of stopping as it should.

Updates to Lidar sensors or the software that interprets the raw data can fix this vulnerability. For example, manufacturers could teach the software to look for telltale signs of decoy reflections added by the laser attack.

“Uncovering this liability allows us to build a more reliable system,” said Yulong Cao, a Michigan doctoral student and lead author of the study. “In our paper, we show that previous defense strategies are insufficient and propose changes that should remedy this weakness.”