I’d hate to be walking down the street and suddenly be affected because my eyes are just “different”
Smith: these are safety glasses, not a facial-recognition firewall.
Will digital cameras improve so that they are less sensitive to lasers? Or will Lidar improve to use less-powerful lasers?
1550 nm lidars for self driving cars are a relatively recent thing with only a handful of companies (AEye, Luminar, Blackmore) making them. The benefit is that, thanks to the extra power, they have longer range. Unfortunately they fry cameras and are very expensive.
It's also worth pointing out that the 1550 nm pulsed lidars like AEye and Luminar with fiber lasers may have much shorter pulses than the 5 ns pulses of, say, a 905 nm Velodyne lidar. So, not only is the average power 1000 times higher, but the peak power may be even higher.
It's generally safe to assume that you're not going to think about this for ten seconds and discover a danger that has been missed by every single person to ever contribute to the exposure limits by thought or by case study.
But then what would HN be for?
Not safe to assume the risks don't exist just because they aren't mentioned in an Ars Technica article.
I would have thought it would be safe to assume a company wouldn't mount a laser on a car that would permanently ruin peoples cameras, but here we are.
I don't believe that the risks don't exist. I believe that what risks exist are extremely unlikely to be something that can be pointed out with a one-line comment on HN. I don't believe that because of this article. I believe that because lasers are unbelievably useful and widely-deployed in industry and tend to cause immediate, visible, and unmistakable damage, so organizations like OSHA have studied them extensively and failures are expensive enough that operators put actual effort into minimizing risk.
> I would have thought it would be safe to assume a
> company wouldn't mount a laser on a car that would
> permanently ruin peoples cameras, but here we are.
I would have thought that that it would be safe to assume that people wouldn't strap 200kW motors to two-ton lumps of metal and send them hurtling around under purely manual control with no physical limits or safety barriers separating them from foot traffic, but here we are.
If the price we pay for eliminating the leading cause of violent death worldwide is that we have to stop pointing cameras at everything, so be it.
> When injury from exposure to microwaves occurs, it usually results from dielectric heating induced in the body. Exposure to microwave radiation can produce cataracts by this mechanism, because the microwave heating denatures proteins in the crystalline lens of the eye (in the same way that heat turns egg whites white and opaque). The lens and cornea of the eye are especially vulnerable because they contain no blood vessels that can carry away heat.
plus are there any weather conditions the light become visible?
edit: > "Cameras are up to 1000x more sensitive to lasers than eyeballs," Dussan wrote. "Occasionally, this can cause thermal damage to a camera's focal plane array." - I'm still not entirely sold.
> Other lidar makers use lasers with a wavelength of 1550nm. This tends to be more expensive because sensors have to be made out of exotic materials like indium-gallium arsenide rather than silicon. But it also has a big advantage: the fluid in the human eye is opaque to 1550nm light, so the light can't reach the retina at the back of the eye. This means lasers can operate at much higher power levels without posing an eye safety risk.
> That energy longer than 1400nm is generally absorbed by the cornea and lens, but it is still energy, and it is not a hard bandpass filter per se. Safety is relative at higher wattages.
sure, but does this hold true for most land-based animals and birds? How much range does this give us? reflections can shift by half a wavelength - is that still opaque too?
I'm not saying this isn't safe, but suddenly having a tonne of high power laser sources pointing everywhere at all times might actually have some consequences still...
(I'm not sure I'm not getting some crossed units when I tried to resolve that into numbers normals like me would understand; is that really 1/e^1000 transmitted per centimeter travelled? e^1000 is a big number. There's even bigger ones on that chart. Then again, if a two-atom-thick layer of gold is enough to make something look like gold and completely obscure what's underneath, I guess that might make sense and my intuition is just off, because when I convert that into this sort of scheme I get big numbers there, too.)
So...it'll just fry the front of the eye?
(I'm rather happy with how well my estimate corresponds to the figure from Wikipedia: "The eye focuses visible and near-infrared light onto the retina. A laser beam can be focused to an intensity on the retina which may be up to 200,000 times higher than at the point where the laser beam enters the eye.")
Also why should I listen to some rando salesman named Dussan? Is he an ophthalmologist?
Another site claims "However, there are no MPEs for sensors such as CMOS or CCD chips. This means a show may be perfectly safe for eyes, but could possibly damage a camera sensor. One reason is that camera lenses may gather more laser light, and concentrate it to a finer point. Another reason is that CMOS or CCD sensors are more easily damaged than the eye."
I've seen f/1.6 lenses, where as the eyes are apparently f/8.3 - f/2.2, so yes the eyes take in less light, but it's substantially harder than an EF-S mount to swapout.
edit: also, what about animals with good night vision? they are more sensitive and take in more light - are they at danger?
Highly doubt this will happen anytime soon. Machine vision has come a long way past decade but nonetheless it is far from being reliable for driving. I think lidar is pretty much a necessity for safety's sakes.
Not even remotely true. LIDAR is used to SLAM in real-time. Think of it as a radar on steroid. For successful reliable autonomous driving you need both machine vision and lidar. If you only use machine vision, you end up like Teslas and slam into stationary fire trucks. So you use machine vision for reading signs, lane localization, etc and lidar to prevent accidents.
> a cross country road trip on any highways.
That's not the agreed upon definition of self driving. Also, you should probably take a look at this: https://en.wikipedia.org/wiki/Self-driving_car#Testing
Waymo on average travels for 5,127.9 miles before a human disengagement is necessary measured over 635,868 miles. Cruise, 5,224 miles. Tesla, 2.9 miles.
> Not you or anyone else can say either machine vision OR lidar are necessary
Well since pretty much every single company that exists right now in the autonomous car space with the exception of Tesla uses LIDAR, I am inclined to think that's the way to do it.
All in all, machine vision is way too unreliable for anything above SAE level 2. Due to the nature of machine learning algorithms used (CNN) you can't be sure that a firetruck will always be recognized as a firetruck. The only way to reliably "see" a firetruck is via lidar/radar.
And the solution is far more simpler than you think the domes for the sensor packages are simply better rated, as they are rated for true 99.999% absorption for both UV and IR (beyond FLIR ranges) some of them also have an inner layer that would darken if the light penetrates that deep creating a dark spot that would block the light completely since replacing a dome is cheaper than replacing a cryogenic IR sensor array.
But beyond that you can blind the sensors with most pen lasers as well as damage it with visible light, the reason why it’s not much of an issue is that due to simple physics you won’t be able to focus a beam narrow enough to hit the sensor even atmospheric refraction alone would prevent that from happening and any laser capable of damaging the sensor at the point would likely be able to damage the drone.
The IR dome protection was mostly implemented to protect the sensor if the drone is painted with an illuminated.
Or just that lasers in general are known to do this and the car's laser probably did it?
I've used band-pass filters which reject everything except a narrow band, so that camera sees just the IR LED (chosen to fall in the middle of passed band) and they cost a few USD in small volumes.
If you went to a Diwali laser show you would get the ultimate camera death mix, fine powder and lasers :-).
I wonder if they had warning signs telling people about the active lasers and possible camera damage. If they didn't, I feel like replacing the camera is just the right thing to do. It was CES, people are bound to have cameras.
Edit: looks like they're using SWIR (~1500nm) so glass/acrylic/polycarbonate will pass it just fine.
 Quanergy Perimeter Security System https://www.youtube.com/watch?v=IFCgEzdrbQM
 Quanergy's LiDAR-based security solution https://www.youtube.com/watch?v=PpYDWb2yX_M
Good to know they claim you can do it with a laser that's not dangerous to the human eye.
You can see the purplish marks just slightly down and left of centre here - https://i.ibb.co/THMS42D/dsc00006.jpg
So, if we define "traversal time" as the time required for the dot of car A to sweep the aperture of car B, in each traversal time there's 1 chance in N^4 of perfect alignment, where N is the ratio of dot size to the full FOV. Maybe it happens once in a blue moon, but if the sensor can withstand full illumination for more than one traversal time, you would start having to compound 1/N^4 events in order to fry a sensor. I wouldn't count on it.
Well, I wouldn't count on that particular mechanism, at least. I'm sure there will be cases where a coating degrades and lets broad-band sunlight in, or a sensor parks on the sun, etc. Sensors will get fried, but not because the engineers making them were too stupid to consider interference.
>Crucially, self-driving cars also rely on conventional cameras. So if those lidars are not camera-safe, it won't just create a headache for people snapping pictures with handheld cameras. Lidar sensors could also damage the cameras on other self-driving cars.
There are military jamming devices that can quite handily permanently damage radars not designed to distinguish it.
I'm happy to be proven wrong though as I wouldn't want something unsafe on the streets as much as anyone else.
Your cell phone camera has a wide angle lens. it's at the longer focal lengths where one might expect sensor damage, or even melt the curtains of shooting directly into the sun when high above horizon.
I guess the answer will be "because it's cheaper to manufacture", but is there any other reason?
Cameras also get regularly damaged at concerts and light shows with visible lasers. Those lasers are class 2 lasers, though, and can output more light while not being blocked by the IR filter in cameras.
This company claims they have a 1550 band laser operating in free space air, spraying all over the place, with a 1000 meter range? Oh, great.
It's not in the visible spectrum so you can't tell if it's damaging your eyeballs, either.
(Or maybe not, owing to the potential liability.)
Assuming you fear death enough to give your full attention to driving safely, how do you hope to accomplish this wonderful utopia? There will always be drunk, distracted, aggressive, tired, epileptic, vision impaired drivers as long as humans are driving. The true solution is to remove all human drivers and only have computers that don't need to worry about unpredictability of other cars, because the cars communicate directly.
And by tech I'm including security, because that's the part that I'm worried about too. There's so much incentive for bad actors to mess with these systems, and the consequences are catastrophic.
Just imagine setting up a hidden 1550nm lidar jammer/spammer to confuse cars on a freeway. You could make cars crash and nobody would even know it was there. Even if it made only 0.1% of the cars just slightly more likely to crash, it could still cause deaths.
Also RCE and DOS attacks now the cars have remote start and other stupid "features" that increase the attack surface.
I'm sure there's many more malicious scenarios you can imagine.
Everyone on here seems so keen for these self-driving cars to take over, but there's just so many avenues for abuse and our road laws are not set up to handle all the complications.
Humans are bad drivers, self driving cars do not have to be very good to be substantially better than humans.
You dont have to buy in to progress if you dont want to. But getting angry about it will only give you a headache.
As I understand it, increased safety is one of the primary motivators for self-driving cars. I think it's fairly obvious that this goal has not been realized yet. But it's one motivator for a lot of people, one which you don't seem to acknowledge.
Society does not require individual people's consent in determining which vehicles drive on the road. If it did, there would be no large pickup trucks, because I would never have allowed them.
Why? Not only is he a genius (citation needed), but he's also highly invested in big-risk companies working toward a healthy future. I wish more people were like Elon.