It's surprisingly easy to "blind" a LiDAR-equipped vehicle.
LiDAR sensors have been hailed as a necessary step towards achieving greater, more widespread adoption of semi-autonomous or autonomous driving. Still, a new study conducted by a trio of universities has demonstrated how the technology's ability to perceive hazards can be easily thwarted.
The study was a joint effort between the University of Florida, the University of Michigan, and the University of Electro-Communications based in Japan. Researchers from these institutions demonstrated that using laser-based spoofing techniques can disrupt LiDAR's perceptive abilities to an alarmingly effective degree, to the point where an autonomous vehicle (AV) will collide with a pedestrian it would generally be able to perceive.
If lasers are directed at self-driving cars, the LiDAR system will be unable to process the necessary data to initiate an evasive maneuver. For example, the vehicle won't begin an emergency braking procedure, making it far more vulnerable to collisions with other cars or pedestrians.
By pointing a laser at the LiDAR camera, the system mistakenly processes this disruptive laser signal instead of the signal that receives reflections from real objects. Hence, the car is unable to identify obstacles.
The result is effectively a blinded LiDAR system and the vehicle losing the required accuracy of its autonomous features. While nobody is likely to point lasers at LiDAR-equipped vehicles, the study reveals a shortcoming in the tech that has, until now, been overlooked.
The researchers conducted a series of experiments to test how much they could disrupt the LiDAR. In one simulation scenario, the AV collided with a pedestrian at around 16 mph. The team said that using the laser-based tool to disrupt the data that the LiDAR system would usually receive is called a Physical Removal Attack (PRA). This demonstration could lead to "severe consequences" and endanger both the AV's occupants and surrounding drivers or pedestrians.
In the researchers' various moving vehicle scenarios, the attacks achieved a success rate of over 90% on numerous occasions for removing 90% of the target pedestrian cloud points. In plain language, the LiDAR system lost its ability to read and interpret 90% of the data it needed to enforce an accurate evasive maneuver for the AV.
These new findings are concerning for automakers like Volvo, which have touted its LiDAR-equipped vehicles as being among the most advanced.
Until now, Tesla has been criticized for its heavy reliance on cameras and not LiDAR to enable its driver-assistance features. Although the cameras alone can't identify objects as accurately as LiDAR, it seems that they may not have the same vulnerabilities to these laser-based attacks as LiDAR-equipped cars like the Volvo EX90. It's not just Volvo employing LiDAR, though, as Mercedes-Benz also beat the competition by having its first Level 3-capable vehicle approved in the USA. The tech will be available for the S-Class and electric EQS.
Like any new technology, LiDAR has its clear advantages but may come with issues that aren't immediately obvious. Before the tech is implemented widely, we hope automakers make changes to address LiDAR's vulnerabilities.
Join The Discussion