> And since we’re asked about it occasionally, the lasers we use are well under the eye safety limit and have been certified as Class 1 eye-safe lasers by third party labs.
Oh geeze. I’ve never even considered the fact that there’s a possibility looking at these cars might cause vision issues!
It might be fun to come up with the probability of a worse case scenario, like parked on the crest of a hill in bumper to bumper traffic.
Although probably not practical to worry about in reality, there is some probability that all 200 cars you could see shine their lasers directly into your eyes at the exact same time. There is some probability of accidentally receiving a momentary higher than safe combined pulse from multiple vehicles.
> Although probably not practical to worry about in reality, there is some probability that all 200 cars you could see shine their lasers directly into your eyes at the exact same time. There is some probability of accidentally receiving a momentary higher than safe combined pulse from multiple vehicles
Would that actually matter? Each of those would be coming from a different direction, and be focused on a different spot on the retina. No particular spot on the retina would receive an unsafe exposure.
A lot of infrared lasers are dangerous because of this. Blink response is part of the reason visible lasers are classified differently than invisible lasers.
Through my work (shameless plug: www.dotproduct3d.com) I've assessed a fair number of "Time of Flight" 3D sensors (which also use flashes of IR light -- but aren't really time-of-flight but that's another story) and, while generally pretty cool, all of them so far have suffered more or less severe multi-path effects. Which is why stereo / structured light or single-point LiDAR are still better for a lot of use cases. I can't wait for these MPI problems to be finally solved.
It should be a smaller problem than with the Kinect. The Kinect is not true time of flight like you said, whereas ouster is. If you know when photons actually arrived you know the first light pulse you receive is the one you care about, and the later longer paths you can ignore. The Kinect conflates these.
So, I don't think it will be a problem for them but it is something they need to consider when designing their DSP.
The reality is somewhat more complicated than that. Signal processing at this level is statistical in nature, so the threshold itself is effectively timing-driven.
Hard to explain in a few words, but the principle of the lock-in amplifier is a good place to start.
I still don't understand how there is no threshold based on amplitude (correlation or not). Something like mist would show up as a noise floor, a speck of dust (that hangs around for long enough) may be one peak, a mirror another, and an object seen in the mirror will be another. You will have multiple peaks that may stay around longer than others and, but there will still be peaks, one of them being the "true" surface.
Compared to other sensors, lidars are not that prone to interference because:
* it only takes 1 microsecond to make a ranging measurement up to 150 m, so your detector is on for a short time
* lasers only illuminate a small spot, and the detector is also looking at a similarly small spot, so it is unlikely for two lidars to point in the same spot
Now, even if it does interfere, you may see a stream of random points pointed towards the interference source. This may happen if, say, you point a lidar directly at the sun, or if you have multiple lidars mounted on the same vehicle. Such random points are easily rejected as outliers and do not affect the vast majority of the scene. Most self driving cars (I hope) should have outlier rejection schemes that deal with outliers caused by this and other sources, such as snow, smoke, and so on.
For this reason we see many self driving cars bristling with a bunch of lidars next to each other with no problems. For example the Cruise/GM ones have five Velodynes on top.
There's a case for adding some jitter, maybe 10us, to the laser timing. That prevents any attacker from synchronizing a jammer. Common technique in military radars. If you can't synch, you can't present an illusion of something being closer than it is, and any return from a real obstacle will come in before a jamming signal from a further away transponder or mirror.
Active systems like radars have a "burn-through" short range - at some short distance, the sensing system overpowers a jammer. So if you're seeing junk at distance, but good signal at short range, you know you're being jammed and have to slow down.
Active jamming is not that effective against things that receive directionally, as people using car "radar jammers" near military bases sometimes discover. They show up on military radars as hostile targets. That's even filtering down to police LIDAR guns.
Velodyne lidars have a phase lock feature so that each lidar is out of phase with each other to prevent interference patterns. This is a bigger issue with sensors that are permanently mounted together but it'll also occur when there are many autonomous vehicles driving together.
Just another of the many edgecases to figure out before autonomous vehicles become mainstream.
I've never had a problem with a bunch of robots in the same room all using horizontal planar lidars. Mostly due to low duty cycles, I suppose, but no sensor is perfect and you've got to have some sort of noise rejection in any event.
When they talk about rejecting solar interference I'd presume, but can't be sure, that whatever they do would also help a lot with lidar on lidar interference.
The interference will most likely show up as noise. Its SPAD detectors have high noise compared to APD so the increased noise due to interference with other lidars would be minimal.
the article is a bit light on real details, but all the affordable LIDAR technology I have seen used Linear-Feedback Shift Registers, such that different devices operating at the same frequency are still on different channels.
It's unclear if this product works in the time-domain or in the frequency domain...
Shorter range lidars (<10m) tend to work in the frequency domain because you can get pretty good ranging accuracy.
Longer range lidars work in the time domain. The SPAD detectors used by Ouster create a pulse when a single photon is detected and so you're measuring the time of flight of the photon being emitted by the vscel and then being detected by the SPAD array.
I don’t think 850nm is a good choice for LIDAR for several reasons. Firstly, the article talks about absorption but that’s not the only phenomenon that they need to be concerned about. The other is scattering and 850nm will scatter from particulate matter in the air (like water vapor in fog) and suffer reduced range.
The 2nd reason why 850nm sources are not a good choice are VCSELs themselves. VCSELs are low power devices and always will be due to the narrow current aperture in their construction, you simply can’t send much current through it because even at low currents the current density is already very high. The 850nm wavelength is also limited in how much output power you can get and still have an eye safe device. Additionally, VCSELs do not work well at high temperatures. Their LIV curve bends over meaning that the combination of reduced slope efficiency and self heating of the junction cause the output power to decrease as you increase current beyond a certain point.
Lastly, VCSELs are relatively noisy devices. VCSELs are not single spatial mode devices and there is mode partition noise caused by the sloshing around of power between the various modes. When combined with RIN, this places an Upper bound on the SNR of the system. Finally, if you are already doing photon counting, you’re pretty much at the limit of this tech already.
To my mind, longer wavelength tech is more promising because it can use higher powers and still be eye safe, has better immunity to scattering in various atmospheric conditions which gives it longer range. Longer wavelengths can leverage technologies like silicon photonics to produce more advanced detectors (coherent detectors which will be more sensitive and improve range). Generally the lasers at these wavelengths are less noisy as they’ve been optimized for decades for use in long distance communication.
As far as I know detection SNR goes way down with longer wavelengths. I think luminartech.com claims that they have specific IP to make this work, but apparently it's a hard problem, otherwise everybody would be doing it already.
Really wish 3d lidar costs would come down. Everyone is designing for car performance and price points it seems. I have a ton of applications I could throw one of these on at 1k$.
That one only has 12 m range. The 2D lidars from Hokuyo, SICK, etc with similar range to the Ouster OS-1 (which has 140 m range) cost thousands of dollars.
Costs are definitely coming down. The OS-1 sensor is 1/6 the price of Velodyne's 64 channel sensor. Ouster has been pretty explicit about our intent to keep driving down prices, and we have significant structural advantages covered in the article that will allow us to do so in the coming months and years.
Yea, my space is mostly industrial so its a more cost sensitive area.
Out of curiosity, this artifact with the van in the below picture common with flash lidars? https://imgur.com/a/OI4UZt9
I am guessing the van is painted with something that strongly absorbs your laser frequencies (black paint). Is this something that your lidar is working to get around?
Yeah - we're actually exploring how to address the industrial market more effectively right now. There's a lot of interest in moving from 2D to 3D lidar.
Part of that is probably just the resolution you're seeing on the point cloud (i.e. if it were up on a 4k OLED you'd see it a lot better than you do on that screenshot) and part of it is definitely that lidar as a broad category has a comparatively harder time with this type of surface than other surfaces. That's why it's super important when comparing lidar units to see what reflectivity they're quoting. 80% vs 10% reflectivity typically reduces range significantly.
PS if you want to get in touch directly, my email is derek.frome at ouster dot io
Nice. Not really here yet, but the semiconductor physics sounds promising. The device they're actually selling is yet another spinning scanner, not a true flash unit.
It's often easier to make a small chip and spin it around on a rotating platform than to make a big chip. For a similar reason, rotating line cameras are a thing for panoramic imaging [1]. As the technology matures and become cheaper, we will see more solid state technology, just as bundles of regular area scan cameras with fisheye lenses have displaced rotating line cameras.
Princeton lightwave is also reportedly doing scanning flash even with 1550nm GmAPDs. Note that spads in InGaAs have much higher dark count rates (i think due to poorer crystal growth technology?)
I'm curious how these systems would function against a malicious entity. For example, if someone flooded the sensor with laser light.
That sort of behavior might be made illegal (if it isn't already), but I'd be curious to know if self-driving systems have ways of dealing with it, or if they just shut down
It's pretty illegal to point lasers at vehicles. Many states and countries have laws against that. For example, in California,
California Penal Code Section 417.27
(c) No person shall direct the beam from a laser pointer directly or indirectly into the eye or eyes of another person or into a moving vehicle with the intent to harass or annoy the other person or the occupants of the moving vehicle.
It is illegal for a laser pointer, where most people defer to the FDA definition of "... handheld lasers that are promoted for pointing out objects or locations."
There is already established case law around this, where larger hobby lasers intended for popping balloons and theatrical lasers are exempt.
You could build a large array of laser emitters intended to obstruct self driving vehicles (say you wanted to create a fake wall to prevent a car from driving the wrong way), and safely ignore these laws.
> It is illegal for a laser pointer, where most people defer to the FDA definition of "... handheld lasers that are promoted for pointing out objects or locations."
No the law in question defines the term, it doesn't defer to the FDA; Cal. Penal Code § 417.27(f): «As used in this section, “laser pointer” has the same meaning as set forth in subdivision (c) of Section 417.25»
Cal. Penal Code § 417.25(c): «As used in this section, “laser pointer” means any hand held laser beam device or demonstration laser product that emits a single point of light amplified by the stimulated emission of radiation that is visible to the human eye.»
I think dynamic patterns can be added to laser beams (similar to what cryptography), which randomly change that will make flooding difficult if not impossible.
Those patterns and various frequency shifts are often used to prevent secondary reflections and to be able to operate in a saturated environment.
A powerful commercial laser can easily damage the CCD but again it’s not an issue drivers can be blinded by lasers already and are constantly blinded by other things like high beams and highway camera flashes the closest I came to being night blinded twice is when a red light camera was triggered on an intersection in both cases I got honked for not seeing the light turn green because I couldn’t see shit for like 5 seconds.
Not every system needs to be resilient to adversaries we don’t drive in armored cars or have active protection systems against ATGMs put on our SUVs.
If an asshole wants to jam cars they’ll be able to they’ll get caught and go to jail that’s more than enough for these systems to be reliable.
If you think about it we already have systems which we rely on everyday that can be jammed with $50 worth of gear like cell phones and GPS these aren’t adversary resistant but we don’t cry about it because it’s not an issue.
It's not an issue because loosing GPS or cell phone coverage is in the vast majority of cases just an inconvenience. Also, no critical system relies on it blindly (part from emergency calls).
A high speed collision is not comparable.
A human has tons of options being blinded by a laser. Obscuring it for one.
Also the ease of which an attack can be performed is of immense importance. It is quite the barrier to pick up a rifle and start to shoot at vehicles. A handheld laser is something people do just for shits and giggles.
This is something that helicopter pilots are quite aware of.
>A human has tons of options being blinded by a laser. Obscuring it for one.
Really? get a 500mw laser into your eyes at highway speeds and I can guarantee you it's a crash, the pain and the shock from being blinded you aren't going to react rationally you going to twitch uncontrollably.
GPS and Cellphones are constantly used for life critical services and can be jammed easily but no one is doing that.
>This is something that helicopter pilots are quite aware of.
Helicopter pilots being blinded by lasers isn't the same thing as drivers at those altitudes the laser beam has spread sufficiently to while remain dangerous at least not being painful.
Helicopters also often come with some sort of auto hover mode these days and you can maintain attitude without sight landing would be hard but you aren't at risk of crashing into anything immediately.
> GPS and Cellphones are constantly used for life critical services and can be jammed easily but no one is doing that.
They are being jammed. Not as often as idiots playing with lasers though. But for silly things like "my employer put a tracking device on my truck, if I jam signals I can do whatever I want".
> Helicopter pilots being blinded by lasers isn't the same thing as drivers at those altitudes the laser beam has spread sufficiently to while remain dangerous at least not being painful.
Can apparently break havoc on sensors though. And this happens over densely populated areas.
> Helicopters also often come with some sort of auto hover mode these days and you can maintain attitude without sight landing would be hard but you aren't at risk of crashing into anything immediately.
I'd guess that if the car's LIDAR is suddenly blinded, the worse that would happen is that the car would slow down and stop. There would probably be recorded data on the incident that might even help find the one who did it.
super neat stuff, I visited their offices recently and was really impressed by what they were doing. From what I can tell, everything was done in house
Laser safety is a well known issue, IEC60825 offers mathematical models to calculate the safety category (e.g class 1/2) of a laser device, which are based on physics models of the eye from first principles and empirical data. There is nothing particularly special about LIDAR lasers that makes these models not valid as far as I know. Generally, there is a tradeoff between exposure time and intensity, so LIDAR systems have redundant hardware methods to ensure the pulse time is not exceeded.
Any laser product generally has to be tested to IEC60825 to check which category it fits within, under the safety rules of the jurisdiction (e.g. CE for EU, FDA for USA). This LIDAR system would have to pass that too, and it should be reasonably easy for them to check if it would when developing it.
The laser isn't tested to IEC60825. Instead the manufacturer (Oester in this case) would submitted information about the laser scanning system design to the FDA to get approved as a Class 1 device.
thanks for mentioning IEC60825, which will come in handy for a project, but I would always feel out of place asking for it on a tangential HN thread, so I am really glad it popped up spontaneously...
1) Does it offer the eye models themselves or only some results of the eye models? If it is only based on eye models, where can I learn more about the eye models/empirical data themselves?
2) Does IEC60825 only pertain to laser safety, or does it also treat LED safety?
The Ouster OS-1 in the article, as well as all other automotive lidars that I know of, are class 1 laser eye-safe, meaning that it is safe even if you put your eye right up to it for hours.
The power also decreases dramatically once you get far away from it, since the laser beams spend most of their time pointed in different directions, and the collimation is not perfect.
Besides, infrared security cameras put out just as much, if not more 850 nm - 940 nm light.
"lasers we use are well under the eye safety limit and have been certified as Class 1 eye-safe lasers by third party labs."
I'm a bit surprised by that since IIRC 850 nm can go through the water in your eye to reach your retina but won't trigger a blink reflex. So it has to be pretty low power to get that rating.
905 nm can also go through the water in your eye (the eye safety limit for 905 nm is maybe 2 times higher than for 850 nm), but Velodyne and many other lidar companies have been using 905 nm for years with no problems. The 2x higher sensitivity of silicon CMOS detectors to 850 nm in comparison to 905 nm makes up for the lower power.
This is a pretty incredible achievement. Be sure to follow the #3 "High quality ambient imagery" link too, if you haven't seen it before:
https://www.ouster.io/blog-posts/2018/8/31/the-camera-is-in-...
Relevant HN discussion about that post: https://news.ycombinator.com/item?id=17896942
> And since we’re asked about it occasionally, the lasers we use are well under the eye safety limit and have been certified as Class 1 eye-safe lasers by third party labs.
Oh geeze. I’ve never even considered the fact that there’s a possibility looking at these cars might cause vision issues!
It might be fun to come up with the probability of a worse case scenario, like parked on the crest of a hill in bumper to bumper traffic.
Although probably not practical to worry about in reality, there is some probability that all 200 cars you could see shine their lasers directly into your eyes at the exact same time. There is some probability of accidentally receiving a momentary higher than safe combined pulse from multiple vehicles.
> Although probably not practical to worry about in reality, there is some probability that all 200 cars you could see shine their lasers directly into your eyes at the exact same time. There is some probability of accidentally receiving a momentary higher than safe combined pulse from multiple vehicles
Would that actually matter? Each of those would be coming from a different direction, and be focused on a different spot on the retina. No particular spot on the retina would receive an unsafe exposure.
There are wavelength that are invisible to human eye, those are pretty safe unless the energy levels are high enough to burn a hole in your retina.
Then there are wavelength that aren’t focuse by the eyes, the only way to get hurt with these if the energy reaches weapon-grade levels.
Actually, some non visible wavelengths are dangerous because they don't make you blink.
A lot of infrared lasers are dangerous because of this. Blink response is part of the reason visible lasers are classified differently than invisible lasers.
Worst I can think of is a peephole in a cheap apartment looking over a turn on a freeway.
Perhaps a fisheye mirror on a big truck.
Looking at them through binoculars (basis for ENOHD safety standard).
A big accuracy problem with flash approaches is multi path interference.
https://www.microsoft.com/en-us/research/wp-content/uploads/...
I wonder if Ouster is susceptible to this. There's no mention of it, but I imagine it will be a bigger problem as they add more lines of resolution.
This.
Through my work (shameless plug: www.dotproduct3d.com) I've assessed a fair number of "Time of Flight" 3D sensors (which also use flashes of IR light -- but aren't really time-of-flight but that's another story) and, while generally pretty cool, all of them so far have suffered more or less severe multi-path effects. Which is why stereo / structured light or single-point LiDAR are still better for a lot of use cases. I can't wait for these MPI problems to be finally solved.
It should be a smaller problem than with the Kinect. The Kinect is not true time of flight like you said, whereas ouster is. If you know when photons actually arrived you know the first light pulse you receive is the one you care about, and the later longer paths you can ignore. The Kinect conflates these.
So, I don't think it will be a problem for them but it is something they need to consider when designing their DSP.
This makes me wonder how any of these systems handle the reflective surfaces of cars.
So you hit a chrome bumper or maybe a glossy black point. The later pulses could arrive at a much higher amplitude.
They don't care about the amplitude, just the timing.
They're measuring the time to some amplitude at the sensor. There must be some amplitude threshold, otherwise would trigger on noise.
The reality is somewhat more complicated than that. Signal processing at this level is statistical in nature, so the threshold itself is effectively timing-driven.
Hard to explain in a few words, but the principle of the lock-in amplifier is a good place to start.
I still don't understand how there is no threshold based on amplitude (correlation or not). Something like mist would show up as a noise floor, a speck of dust (that hangs around for long enough) may be one peak, a mirror another, and an object seen in the mirror will be another. You will have multiple peaks that may stay around longer than others and, but there will still be peaks, one of them being the "true" surface.
I don't see how there is no threshold.
What happens when multiple lidar devices start shooting laser beams at a scene? Will there be interference?
Compared to other sensors, lidars are not that prone to interference because:
* it only takes 1 microsecond to make a ranging measurement up to 150 m, so your detector is on for a short time
* lasers only illuminate a small spot, and the detector is also looking at a similarly small spot, so it is unlikely for two lidars to point in the same spot
Now, even if it does interfere, you may see a stream of random points pointed towards the interference source. This may happen if, say, you point a lidar directly at the sun, or if you have multiple lidars mounted on the same vehicle. Such random points are easily rejected as outliers and do not affect the vast majority of the scene. Most self driving cars (I hope) should have outlier rejection schemes that deal with outliers caused by this and other sources, such as snow, smoke, and so on.
For this reason we see many self driving cars bristling with a bunch of lidars next to each other with no problems. For example the Cruise/GM ones have five Velodynes on top.
There's a case for adding some jitter, maybe 10us, to the laser timing. That prevents any attacker from synchronizing a jammer. Common technique in military radars. If you can't synch, you can't present an illusion of something being closer than it is, and any return from a real obstacle will come in before a jamming signal from a further away transponder or mirror.
Active systems like radars have a "burn-through" short range - at some short distance, the sensing system overpowers a jammer. So if you're seeing junk at distance, but good signal at short range, you know you're being jammed and have to slow down.
Active jamming is not that effective against things that receive directionally, as people using car "radar jammers" near military bases sometimes discover. They show up on military radars as hostile targets. That's even filtering down to police LIDAR guns.
Velodyne lidars have a phase lock feature so that each lidar is out of phase with each other to prevent interference patterns. This is a bigger issue with sensors that are permanently mounted together but it'll also occur when there are many autonomous vehicles driving together.
Just another of the many edgecases to figure out before autonomous vehicles become mainstream.
Cruise bought Strobe a while ago, I wonder how that's going.
https://www.theverge.com/2017/10/9/16448306/gm-cruise-strobe...
(Disclaimer, I work for GM, but not on any of this)
I've never had a problem with a bunch of robots in the same room all using horizontal planar lidars. Mostly due to low duty cycles, I suppose, but no sensor is perfect and you've got to have some sort of noise rejection in any event.
When they talk about rejecting solar interference I'd presume, but can't be sure, that whatever they do would also help a lot with lidar on lidar interference.
The interference will most likely show up as noise. Its SPAD detectors have high noise compared to APD so the increased noise due to interference with other lidars would be minimal.
the article is a bit light on real details, but all the affordable LIDAR technology I have seen used Linear-Feedback Shift Registers, such that different devices operating at the same frequency are still on different channels.
It's unclear if this product works in the time-domain or in the frequency domain...
> light on real details
If you want more details, Image Sensors World [1] has coverage with more details and links to two patents.
> It's unclear if this product works in the time-domain
It is a pulsed time of flight direct detection lidar, not a modulated continuous wave lidar.
[1] http://image-sensors-world.blogspot.com/2018/11/ouster-discu...
thanks, that is helpful!
Shorter range lidars (<10m) tend to work in the frequency domain because you can get pretty good ranging accuracy.
Longer range lidars work in the time domain. The SPAD detectors used by Ouster create a pulse when a single photon is detected and so you're measuring the time of flight of the photon being emitted by the vscel and then being detected by the SPAD array.
I don’t think 850nm is a good choice for LIDAR for several reasons. Firstly, the article talks about absorption but that’s not the only phenomenon that they need to be concerned about. The other is scattering and 850nm will scatter from particulate matter in the air (like water vapor in fog) and suffer reduced range.
The 2nd reason why 850nm sources are not a good choice are VCSELs themselves. VCSELs are low power devices and always will be due to the narrow current aperture in their construction, you simply can’t send much current through it because even at low currents the current density is already very high. The 850nm wavelength is also limited in how much output power you can get and still have an eye safe device. Additionally, VCSELs do not work well at high temperatures. Their LIV curve bends over meaning that the combination of reduced slope efficiency and self heating of the junction cause the output power to decrease as you increase current beyond a certain point.
Lastly, VCSELs are relatively noisy devices. VCSELs are not single spatial mode devices and there is mode partition noise caused by the sloshing around of power between the various modes. When combined with RIN, this places an Upper bound on the SNR of the system. Finally, if you are already doing photon counting, you’re pretty much at the limit of this tech already.
To my mind, longer wavelength tech is more promising because it can use higher powers and still be eye safe, has better immunity to scattering in various atmospheric conditions which gives it longer range. Longer wavelengths can leverage technologies like silicon photonics to produce more advanced detectors (coherent detectors which will be more sensitive and improve range). Generally the lasers at these wavelengths are less noisy as they’ve been optimized for decades for use in long distance communication.
As far as I know detection SNR goes way down with longer wavelengths. I think luminartech.com claims that they have specific IP to make this work, but apparently it's a hard problem, otherwise everybody would be doing it already.
Really wish 3d lidar costs would come down. Everyone is designing for car performance and price points it seems. I have a ton of applications I could throw one of these on at 1k$.
If you don't need 3D, 2D LIDAR is pretty cheap right now: https://www.adafruit.com/product/4010
That one only has 12 m range. The 2D lidars from Hokuyo, SICK, etc with similar range to the Ouster OS-1 (which has 140 m range) cost thousands of dollars.
Costs are definitely coming down. The OS-1 sensor is 1/6 the price of Velodyne's 64 channel sensor. Ouster has been pretty explicit about our intent to keep driving down prices, and we have significant structural advantages covered in the article that will allow us to do so in the coming months and years.
Yea, my space is mostly industrial so its a more cost sensitive area.
Out of curiosity, this artifact with the van in the below picture common with flash lidars? https://imgur.com/a/OI4UZt9
I am guessing the van is painted with something that strongly absorbs your laser frequencies (black paint). Is this something that your lidar is working to get around?
Yeah - we're actually exploring how to address the industrial market more effectively right now. There's a lot of interest in moving from 2D to 3D lidar.
Part of that is probably just the resolution you're seeing on the point cloud (i.e. if it were up on a 4k OLED you'd see it a lot better than you do on that screenshot) and part of it is definitely that lidar as a broad category has a comparatively harder time with this type of surface than other surfaces. That's why it's super important when comparing lidar units to see what reflectivity they're quoting. 80% vs 10% reflectivity typically reduces range significantly.
PS if you want to get in touch directly, my email is derek.frome at ouster dot io
Nice. Not really here yet, but the semiconductor physics sounds promising. The device they're actually selling is yet another spinning scanner, not a true flash unit.
It's often easier to make a small chip and spin it around on a rotating platform than to make a big chip. For a similar reason, rotating line cameras are a thing for panoramic imaging [1]. As the technology matures and become cheaper, we will see more solid state technology, just as bundles of regular area scan cameras with fisheye lenses have displaced rotating line cameras.
[1] https://en.wikipedia.org/wiki/Rotating_line_camera
What are your thoughts on Geiger LIDAR, ie Princeton Lightwave (which was acquired by Ford/Argo)?
SPADs (single photon avalanche photodiodes) are APD's operated in Geiger mode...
Princeton lightwave is also reportedly doing scanning flash even with 1550nm GmAPDs. Note that spads in InGaAs have much higher dark count rates (i think due to poorer crystal growth technology?)
I'm curious how these systems would function against a malicious entity. For example, if someone flooded the sensor with laser light.
That sort of behavior might be made illegal (if it isn't already), but I'd be curious to know if self-driving systems have ways of dealing with it, or if they just shut down
It's pretty illegal to point lasers at vehicles. Many states and countries have laws against that. For example, in California,
California Penal Code Section 417.27
(c) No person shall direct the beam from a laser pointer directly or indirectly into the eye or eyes of another person or into a moving vehicle with the intent to harass or annoy the other person or the occupants of the moving vehicle.
It is illegal for a laser pointer, where most people defer to the FDA definition of "... handheld lasers that are promoted for pointing out objects or locations."
There is already established case law around this, where larger hobby lasers intended for popping balloons and theatrical lasers are exempt.
You could build a large array of laser emitters intended to obstruct self driving vehicles (say you wanted to create a fake wall to prevent a car from driving the wrong way), and safely ignore these laws.
> It is illegal for a laser pointer, where most people defer to the FDA definition of "... handheld lasers that are promoted for pointing out objects or locations."
No the law in question defines the term, it doesn't defer to the FDA; Cal. Penal Code § 417.27(f): «As used in this section, “laser pointer” has the same meaning as set forth in subdivision (c) of Section 417.25»
Cal. Penal Code § 417.25(c): «As used in this section, “laser pointer” means any hand held laser beam device or demonstration laser product that emits a single point of light amplified by the stimulated emission of radiation that is visible to the human eye.»
If you read the definition you just quoted at me, it is basically the same.
Any device that emits more than a single point of light, or is not hand held is not covered.
You're right, I stand corrected. It is just a matter of time before that gets banned as well, though.
How do drivers currently handle things if someone blinds them with a laser?
Probably similar laws and penalties as if you flooded a drivers eyes with lasers.
Relevant xkcd:
https://www.xkcd.com/1958/
They could be using coding techniques like a Barker code or something
I think dynamic patterns can be added to laser beams (similar to what cryptography), which randomly change that will make flooding difficult if not impossible.
Those patterns and various frequency shifts are often used to prevent secondary reflections and to be able to operate in a saturated environment.
A powerful commercial laser can easily damage the CCD but again it’s not an issue drivers can be blinded by lasers already and are constantly blinded by other things like high beams and highway camera flashes the closest I came to being night blinded twice is when a red light camera was triggered on an intersection in both cases I got honked for not seeing the light turn green because I couldn’t see shit for like 5 seconds.
Not every system needs to be resilient to adversaries we don’t drive in armored cars or have active protection systems against ATGMs put on our SUVs.
If an asshole wants to jam cars they’ll be able to they’ll get caught and go to jail that’s more than enough for these systems to be reliable.
If you think about it we already have systems which we rely on everyday that can be jammed with $50 worth of gear like cell phones and GPS these aren’t adversary resistant but we don’t cry about it because it’s not an issue.
It's not an issue because loosing GPS or cell phone coverage is in the vast majority of cases just an inconvenience. Also, no critical system relies on it blindly (part from emergency calls).
A high speed collision is not comparable.
A human has tons of options being blinded by a laser. Obscuring it for one.
Also the ease of which an attack can be performed is of immense importance. It is quite the barrier to pick up a rifle and start to shoot at vehicles. A handheld laser is something people do just for shits and giggles.
This is something that helicopter pilots are quite aware of.
>A human has tons of options being blinded by a laser. Obscuring it for one.
Really? get a 500mw laser into your eyes at highway speeds and I can guarantee you it's a crash, the pain and the shock from being blinded you aren't going to react rationally you going to twitch uncontrollably.
GPS and Cellphones are constantly used for life critical services and can be jammed easily but no one is doing that.
>This is something that helicopter pilots are quite aware of.
Helicopter pilots being blinded by lasers isn't the same thing as drivers at those altitudes the laser beam has spread sufficiently to while remain dangerous at least not being painful.
Helicopters also often come with some sort of auto hover mode these days and you can maintain attitude without sight landing would be hard but you aren't at risk of crashing into anything immediately.
> Really?
Yes.
> GPS and Cellphones are constantly used for life critical services and can be jammed easily but no one is doing that.
They are being jammed. Not as often as idiots playing with lasers though. But for silly things like "my employer put a tracking device on my truck, if I jam signals I can do whatever I want".
> Helicopter pilots being blinded by lasers isn't the same thing as drivers at those altitudes the laser beam has spread sufficiently to while remain dangerous at least not being painful.
Can apparently break havoc on sensors though. And this happens over densely populated areas.
> Helicopters also often come with some sort of auto hover mode these days and you can maintain attitude without sight landing would be hard but you aren't at risk of crashing into anything immediately.
Good luck doing that downtown.
I'd guess that if the car's LIDAR is suddenly blinded, the worse that would happen is that the car would slow down and stop. There would probably be recorded data on the incident that might even help find the one who did it.
super neat stuff, I visited their offices recently and was really impressed by what they were doing. From what I can tell, everything was done in house
Is something like phased array lidar useful? The wavelength is so small I could see how the interference patterns might not be useful
How dangerous are these lasers?
If hundreds of cars will have them, your eyes will be scanned multiple times per day.
Laser safety is a well known issue, IEC60825 offers mathematical models to calculate the safety category (e.g class 1/2) of a laser device, which are based on physics models of the eye from first principles and empirical data. There is nothing particularly special about LIDAR lasers that makes these models not valid as far as I know. Generally, there is a tradeoff between exposure time and intensity, so LIDAR systems have redundant hardware methods to ensure the pulse time is not exceeded.
Any laser product generally has to be tested to IEC60825 to check which category it fits within, under the safety rules of the jurisdiction (e.g. CE for EU, FDA for USA). This LIDAR system would have to pass that too, and it should be reasonably easy for them to check if it would when developing it.
The laser isn't tested to IEC60825. Instead the manufacturer (Oester in this case) would submitted information about the laser scanning system design to the FDA to get approved as a Class 1 device.
thanks for mentioning IEC60825, which will come in handy for a project, but I would always feel out of place asking for it on a tangential HN thread, so I am really glad it popped up spontaneously...
1) Does it offer the eye models themselves or only some results of the eye models? If it is only based on eye models, where can I learn more about the eye models/empirical data themselves?
2) Does IEC60825 only pertain to laser safety, or does it also treat LED safety?
Not very dangerous.
The Ouster OS-1 in the article, as well as all other automotive lidars that I know of, are class 1 laser eye-safe, meaning that it is safe even if you put your eye right up to it for hours.
The power also decreases dramatically once you get far away from it, since the laser beams spend most of their time pointed in different directions, and the collimation is not perfect.
Besides, infrared security cameras put out just as much, if not more 850 nm - 940 nm light.
"lasers we use are well under the eye safety limit and have been certified as Class 1 eye-safe lasers by third party labs."
I'm a bit surprised by that since IIRC 850 nm can go through the water in your eye to reach your retina but won't trigger a blink reflex. So it has to be pretty low power to get that rating.
905 nm can also go through the water in your eye (the eye safety limit for 905 nm is maybe 2 times higher than for 850 nm), but Velodyne and many other lidar companies have been using 905 nm for years with no problems. The 2x higher sensitivity of silicon CMOS detectors to 850 nm in comparison to 905 nm makes up for the lower power.