Jim Park at Truckinginfo.com wrote an interesting story questioning the safety of lidar systems. Some types of lidar could potentially cause damage to human eyes, while other types could be hurting cameras that other autonomous depend on for safe operation (not to mention cameras used by traffic equipment, cell phones, and professional photography).
Lidar uses light (from lasers) to get a picture of its surroundings the same way that radar uses radio waves. By sweeping around with a laser that fires weak pulses of light and then seeing what reflections come back, an autonomous vehicle can “see” other vehicles, buildings, pedestrians, and other things that it must be aware of on the road. Tesla doesn’t use lidar for autopilot or the in-development Full Self Driving option, instead depending on cameras alone for computer vision. Most other organizations use lidar, and often have unsightly spinning “salad shooter” type things on the vehicle’s roof and sides.
The possible safety problem comes from the ability of a laser to damage eyes and cameras. A common lidar frequency is 905 nanometers. At that frequency, too much power can temporarily or even permanently blind a human eye. For that reason, governments have set power limits to keep them below what theoretically can harm us.
At 1550 nanometers, it’s much more difficult for lidar to penetrate the human eye. The liquid inside our eyes is mostly opaque to those frequencies, and while a very powerful laser could hurt the surface of your eye, it probably can’t permanently blind you by burning your retina. “The ocular fluid is largely clear at visible wavelengths and out to about 1,300 or 1,400 nm but becomes nearly opaque (think welder’s glass) at 1,550 nm,” said Jeff Hecht, a laser expert, to Truckinginfo.com.
“On the other hand, 905-nm light does reach the retina and could cause eye damage. I wouldn’t want one of those in my car or truck.” he said.
Autonomous vehicle companies are starting to use the 1550 nanometer frequency because government power limits are 1000 times higher. This allows for lidar that gets much better range and allows AVs to “see” well with the vehicle traveling at highway speeds. It’s akin to not outdriving your headlights at night. To go faster, AVs using lidar need brighter lidar.
Even if lidar doesn’t hurt human eyes, it can cause problems for cameras. In one 2019 incident, a journalist took a photo of a vehicle equipped with lidar units. In every subsequent photo, his camera’s sensor showed signs of damage in exactly the same spot. “I noticed that all my pictures were having that spot,” The man told Ars Technica. “I covered up the camera with the lens cap and the spots are there—it’s burned into the sensor.”
It’s well established that lasers can destroy camera sensors, as seen in the below video:
In the CES incident, the lidar company offered to buy the journalist a new camera and promised to take mitigation measures to protect cameras from harm, but not all makers of lidar have promised to do the same nor have they disclosed any measures taken. It’s also been debated whether the CES incident was a fluke or a real problem with lidar systems.
Jeff Hecht, the expert that Truckinginfo.com spoke with, has his own article on the topic. The assumption that 1550 nanometer lasers are safer at much higher power levels is questionable, he says. While the fluid of the eye blocks out light at those frequencies, there is still some risk to the cornea and lens in some situations, so he argues that they can’t really be classified as eye-safe. He also says that there has been at least one incident of corneal damage caused by a 1550 nanometer lidar unit. The company involved refuses to release more information on the incident.
He also points out that cheaper versions of lidar units could leak some light in the 900 nm band, where the light could damage human retinas. Good filters and good construction prevent this, but a damaged or cheaply-built lidar unit could permanently blind people and damage cameras.
As we know from past experience in the automotive world, many “me too” companies will likely emerge in the coming years building cheaper lidar units. If they cut safety corners, even a little, people could be harmed.
This Might Be A Big Deal
Before I explore why this could be such a big problem, keep in mind what limited data I’ve presented here.
All I’ve found so far about these dangers was the ideas of one expert in the field. Needless to say, that’s not enough to make for conclusive proof that lidar units are dangerous. Unfortunately, the technology hasn’t been widely deployed in the real world and not many studies have been done to determine all of the potential risks.
That having been said, if there are real risks to lidar systems, this could cause a lot of problems.
Obviously damage to even a few people’s eyes is a big problem. I get that autonomous vehicles could save a lot of lives by removing the dumbest drivers from the road, but even if there’s a net benefit, the people losing sight occasionally to lidar (when widely deployed) will definitely be on TV telling their terrifying story. This would be bad for every company using lidar for autonomous vehicles, and possibly for AV companies that don’t.
Damage to cameras would be more tolerable for society in many ways because it doesn’t terrify people the way a small chance of losing sight does. That doesn’t mean that damage to cameras, even occasionally, would be a good thing. If there are a lot of vehicles on the road relying on cameras, lidar, or both, blinding a few cameras here and there would be a problem. At worst, a blinded autonomous vehicle could kill people. At best, it would be a nuisance to have to replace expensive cameras from time to time.
Nobody would shed a tear for burnt up traffic enforcement cameras, but other cameras that regulate traffic light timing, monitor roads for conditions, and help with security would definitely all be missed if they went out of order.
This is definitely an issue that needs further scrutiny.
Featured image: Lidar imagery from the National Oceanographic and Atmospheric Administration (Federal agency, public domain)