I was expecting flash LIDAR and MEMS mirror systems to dominate self-driving cars by now, but rotating machinery is still dominant in the US.
The trend in China and Japan is a long-range forward-facing LIDAR coupled with three shorter-range units for side and rear coverage.[1] The long-range unit still costs around US$10,000.
This should come down with volume.
Non-scanning means your flash emits the equivalent of all the beams at the same time. Look at their flagship spec, you must master 160A next to the other electronics. Two board designs are more expensive and error prone, can't get 160 amps through a flex-cable easily. Then there is the opening angle, you need wide _and_ long, at least for front and rear facing and the optics for that are challenging. Another influence certainly is that the automated driving craze has been superseded by the AI craze and scaling won't come in the time frame that was predicted some years ago. Disclaimer: worked in AD since 2012 (too late for the urban challenge unfortunately ;)) and in a company building sensors and a full stack.
P.S.: a QNX desktop is possible and actually alive again, but company politics... :/
Also there's the simple physics view of the same problem. The advantage of scanning is that you can focus all the laser pulse energy into one narrow beam. Non scanning means covering the whole field of view at once with that same laser pulse. Then you have a choice. Either somehow deal with the exponentially weaker return pulse (since it's spread over the whole field of view), or try to increase the pulse energy (and there you're limited by laser safety regulations)
From a physics point of view, continuous transmission and correlation detection of a temporally- and spatially-diverse optical signal over the entire field of view addresses the problem of energy being spread out, and gets you better, more robust depth information for less emitted energy than scanning or full-field flashes.
But from an optical and electronics point of view, it's much harder to process the return signal that way, and probably uses a lot more energy due to the processing required (with current tech).
Actually the same rule as for RADAR also applies to LIDAR: single pulse has better energy efficiency, but requires untenable peak power at any vaguely-state-of-the-art signal quality/reception performance levels.
The reason is that you can time-gate the noise out that would otherwise be hitting your correlation accumulators if you have a vague idea of the supposed delay/ToF for the pulse.
However, once you add mechanical scanning, at least for systems with not that many orders of magnitude between range resolution and maximum detection range, you can use systems like mode-locked lasers that for example have around 0.1% native duty cycle, circumvent the issue of peak power through the aperture/scanning 's spatial focusing (each pixel only needs a managable amount of energy, and delivering that in a single pulse won't require unreasonable peak power levels), and still get all the energy-efficiency benefits of single-pulse ranging vs. spread-spectrum/correlation ranging.
The only but major downside is the requirement of mechanical scanning.
Fourth power of the distance, actually. That's the radar equation. You have inverse-square losses going out, as the beam expands with distance. Then you have inverse-square losses coming back when the target is much smaller than the beam. That's the problem flash LIDARs face. It can be overcome with enough laser power out to 20-30 meters.
That's where the beam diameter at the target is much larger than the target, as for aircraft. With a small scanning dot from a LIDAR and a nice big target like a car, almost all the power hits the target, but you still have inverse square losses coming back.
true. My original was just a quick jote on a phone sipping a coffee on Sunday. I admit I simply didn't want to go into the whole "square FOV for the sensor vs. one detector / diode and that combined with the time of flight loss over distance", so I just used "exponential" to mean "it loses power pretty quickly". Apologies for the sloppiness on my part.
Second part of the comment I omitted is was what You mentioned in the beginning. Those 20-30 meters of practical range is why we keep seeing small LIDAR sensors on things like iPhones / iPads (though there I believe the range is even a bit shorter due to the size / power constraints), but not really much beyond that.
For practical demo of what's currently available at the high end of solid state LIDAR (albeit at 40k+ USD), I'd suggest looking at Leica and their BLK2GO PULSE (solid state) vs the rest of the BLK line (rotating laser spot).
For how long? Can't be continuous. What's average power. Yes, flash LIDAR has a power problem. How does their long-range LIDAR work?
> Another influence certainly is that the automated driving craze has been superseded by the AI craze and scaling won't come in the time frame that was predicted some years ago.
Huh. Good point. Want to think about that one.
We might max out at most taxis being self-driving, because that works and sells, but not make it to personal vehicles.
> P.S.: a QNX desktop is possible and actually alive again, but company politics... :/
I'm out of that now. The closed source/open souce/closed source/open source/closed source transitions angered too many people. There was once Netscape/Firefox for QNX.
While i was reading this article i did find another article (AD) that was showcasing a product of LidAR tech in a small package. It seems unpractical to be very useful a long sim module with its own processing unit and on board electronics. The require space for the module to be installed was above the windshield and the roof. Much like other products of the same category, however this was a lot more compact. Seems to be a push for further compact innovation in this category. i was under the impression there was no need for multiple LIdar module on one vehicle, all you needed was one along with cameras to further assist the LIdar module .I do know that under normal operation the LIdar and camera system are separate and have different responsibilities.
Solid state LADAR with 8km range and spectroscopic chemical analysis capabilities have been around for decades (mostly CO2 and Nitrate concentration along with 3D point cloud data etc.)
However, the chances anyone will see that technology in a consumer product is very low. These were also never cost effective, and priced several times more than most cars. Additionally, like all optics these couldn't handle excessive dust, rain, direct sunlight, and bug guts.
Most platforms included millimeter Radar for when vision and LIDAR/LADAR optics fail. QNX simply missed its largest market launch window in the 1990s, and is no longer the path forward for a lot of projects. Note "AI" might be real someday (unlikely an LLM), but every hype-cycle needs to run its course. =3
The em4 does not cost 10000 dollars. It is going to sell for well under 1000$. Hesai's low end ATX is supposed going to be less than 300$, and the AT128 already sells for 400-500$.
It's mainly the laser itself that is the expensive part. If you only care about resolution it's easy, you just need a single-mode laser. But if you care about accuracy it's very difficult, because then the wavelength needs to be stable, and that requires a much more expensive laser. Most people looking for an interferometer are interested in accuracy, unless they're just measuring vibrations.
You can get pretty far with cheap diodes + current and temperature control. Unless you need coherence lengths in the meters range you can make do with cheaper lasers.
While not an integrated system itself, looking for convenient information on the matter I stumbled upon https://arxiv.org/abs/2011.05313 .
As for integrated electro-optical hardware that does coherent optics, I'd like to point at SFP optics; the ones that are sufficiently non-fancy with their electronics (up to 10km single-mode optics; no high-end-multimode optics; no exotic coherent optics) should not have a retimer on the RX side. Once you go to 1-Gbit/s transceivers you'll find ones with a linear RX stage; those should be suitable for rigging simple homodyne interferometry by just feeding a little TX back to the RX taking care to let the RX AGC run the RX path at high gain.
They are available cheaply. You can purchase a blu ray reader and hack it. The reason a ready packaged sensor is not available cheaply is simply because of economics and market size.
ToF sensors have huge applications. Initially ToF sensors also cost thousands of dollars. What huge applications do you see that should have driven the prices of over the counter integrable interferometers down? As I stated blu ray (and DVD, CD and laser disc before it) readers are tiny purpose build inferometers. And they are cheap.
Since when are CD reader optics interferometers?
The linked video shows a CW homodyne LIDAR used for measuring vibration frequency and counting vibration amplitude in fringes.
Last I looked CD readers used a 4-detector sensor's differential low-pass signals for closed-loop track-following so the rotation need not be optically centered.
I also see no reason why optical disc readouts would need homodyne let alone heterodyne readout.
The pits and valleys are spaced λ/4 apart. The reason there is such a stark difference in intensity is because this λ/4 spacing causes interference at pit->valley and valley->pit transition points. Of course this not a standard interferometer but rather a purpose build one. Wikipedia says:
> Interferometry is a technique which uses the interference of superimposed waves to extract information.
And a blu ray player directly uses interference of superimposed waves to extract information. It squarely fits in the homodyne category.
The first table is already outdated, there are COTS units like the STL-27L that operate on dTOF with a 25m range, return thousands of points per sweep, and fit into a package with a footprint the size of a typical wristwatch. Bulky optics are no longer a requirement, and that's even on the very low end of the price spectrum, more expensive units can do 30+ m in a similarly sized form factor.
I was expecting flash LIDAR and MEMS mirror systems to dominate self-driving cars by now, but rotating machinery is still dominant in the US.
The trend in China and Japan is a long-range forward-facing LIDAR coupled with three shorter-range units for side and rear coverage.[1] The long-range unit still costs around US$10,000. This should come down with volume.
[1] https://www.robosense.ai/en/news-show-1908
[2] https://openelab.io/products/robosense-em4-thousand-beam-lon...
Non-scanning means your flash emits the equivalent of all the beams at the same time. Look at their flagship spec, you must master 160A next to the other electronics. Two board designs are more expensive and error prone, can't get 160 amps through a flex-cable easily. Then there is the opening angle, you need wide _and_ long, at least for front and rear facing and the optics for that are challenging. Another influence certainly is that the automated driving craze has been superseded by the AI craze and scaling won't come in the time frame that was predicted some years ago. Disclaimer: worked in AD since 2012 (too late for the urban challenge unfortunately ;)) and in a company building sensors and a full stack. P.S.: a QNX desktop is possible and actually alive again, but company politics... :/
Also there's the simple physics view of the same problem. The advantage of scanning is that you can focus all the laser pulse energy into one narrow beam. Non scanning means covering the whole field of view at once with that same laser pulse. Then you have a choice. Either somehow deal with the exponentially weaker return pulse (since it's spread over the whole field of view), or try to increase the pulse energy (and there you're limited by laser safety regulations)
From a physics point of view, continuous transmission and correlation detection of a temporally- and spatially-diverse optical signal over the entire field of view addresses the problem of energy being spread out, and gets you better, more robust depth information for less emitted energy than scanning or full-field flashes.
But from an optical and electronics point of view, it's much harder to process the return signal that way, and probably uses a lot more energy due to the processing required (with current tech).
Actually the same rule as for RADAR also applies to LIDAR: single pulse has better energy efficiency, but requires untenable peak power at any vaguely-state-of-the-art signal quality/reception performance levels.
The reason is that you can time-gate the noise out that would otherwise be hitting your correlation accumulators if you have a vague idea of the supposed delay/ToF for the pulse.
However, once you add mechanical scanning, at least for systems with not that many orders of magnitude between range resolution and maximum detection range, you can use systems like mode-locked lasers that for example have around 0.1% native duty cycle, circumvent the issue of peak power through the aperture/scanning 's spatial focusing (each pixel only needs a managable amount of energy, and delivering that in a single pulse won't require unreasonable peak power levels), and still get all the energy-efficiency benefits of single-pulse ranging vs. spread-spectrum/correlation ranging.
The only but major downside is the requirement of mechanical scanning.
Quadratically weaker, not exponentially.
Fourth power of the distance, actually. That's the radar equation. You have inverse-square losses going out, as the beam expands with distance. Then you have inverse-square losses coming back when the target is much smaller than the beam. That's the problem flash LIDARs face. It can be overcome with enough laser power out to 20-30 meters.
That's where the beam diameter at the target is much larger than the target, as for aircraft. With a small scanning dot from a LIDAR and a nice big target like a car, almost all the power hits the target, but you still have inverse square losses coming back.
true. My original was just a quick jote on a phone sipping a coffee on Sunday. I admit I simply didn't want to go into the whole "square FOV for the sensor vs. one detector / diode and that combined with the time of flight loss over distance", so I just used "exponential" to mean "it loses power pretty quickly". Apologies for the sloppiness on my part.
Second part of the comment I omitted is was what You mentioned in the beginning. Those 20-30 meters of practical range is why we keep seeing small LIDAR sensors on things like iPhones / iPads (though there I believe the range is even a bit shorter due to the size / power constraints), but not really much beyond that.
For practical demo of what's currently available at the high end of solid state LIDAR (albeit at 40k+ USD), I'd suggest looking at Leica and their BLK2GO PULSE (solid state) vs the rest of the BLK line (rotating laser spot).
> 160A
For how long? Can't be continuous. What's average power. Yes, flash LIDAR has a power problem. How does their long-range LIDAR work?
> Another influence certainly is that the automated driving craze has been superseded by the AI craze and scaling won't come in the time frame that was predicted some years ago.
Huh. Good point. Want to think about that one.
We might max out at most taxis being self-driving, because that works and sells, but not make it to personal vehicles.
> P.S.: a QNX desktop is possible and actually alive again, but company politics... :/
I'm out of that now. The closed source/open souce/closed source/open source/closed source transitions angered too many people. There was once Netscape/Firefox for QNX.
While i was reading this article i did find another article (AD) that was showcasing a product of LidAR tech in a small package. It seems unpractical to be very useful a long sim module with its own processing unit and on board electronics. The require space for the module to be installed was above the windshield and the roof. Much like other products of the same category, however this was a lot more compact. Seems to be a push for further compact innovation in this category. i was under the impression there was no need for multiple LIdar module on one vehicle, all you needed was one along with cameras to further assist the LIdar module .I do know that under normal operation the LIdar and camera system are separate and have different responsibilities.
Solid state LADAR with 8km range and spectroscopic chemical analysis capabilities have been around for decades (mostly CO2 and Nitrate concentration along with 3D point cloud data etc.)
However, the chances anyone will see that technology in a consumer product is very low. These were also never cost effective, and priced several times more than most cars. Additionally, like all optics these couldn't handle excessive dust, rain, direct sunlight, and bug guts.
Most platforms included millimeter Radar for when vision and LIDAR/LADAR optics fail. QNX simply missed its largest market launch window in the 1990s, and is no longer the path forward for a lot of projects. Note "AI" might be real someday (unlikely an LLM), but every hype-cycle needs to run its course. =3
The em4 does not cost 10000 dollars. It is going to sell for well under 1000$. Hesai's low end ATX is supposed going to be less than 300$, and the AT128 already sells for 400-500$.
Slightly offtopic, why is it so difficult to find a cheap and compact laser interferometer that can do sub-micron measurements?
This guy gets close:
https://www.youtube.com/watch?v=MUdro-6u2Zg&t=770s
But why isn't something cheap and small like this commercially available as an integrated system?
It's mainly the laser itself that is the expensive part. If you only care about resolution it's easy, you just need a single-mode laser. But if you care about accuracy it's very difficult, because then the wavelength needs to be stable, and that requires a much more expensive laser. Most people looking for an interferometer are interested in accuracy, unless they're just measuring vibrations.
Can't you solve the stable wavelength issue by using a beamsplitter and a separate reference arm?
You can get pretty far with cheap diodes + current and temperature control. Unless you need coherence lengths in the meters range you can make do with cheaper lasers.
While not an integrated system itself, looking for convenient information on the matter I stumbled upon https://arxiv.org/abs/2011.05313 . As for integrated electro-optical hardware that does coherent optics, I'd like to point at SFP optics; the ones that are sufficiently non-fancy with their electronics (up to 10km single-mode optics; no high-end-multimode optics; no exotic coherent optics) should not have a retimer on the RX side. Once you go to 1-Gbit/s transceivers you'll find ones with a linear RX stage; those should be suitable for rigging simple homodyne interferometry by just feeding a little TX back to the RX taking care to let the RX AGC run the RX path at high gain.
They are available cheaply. You can purchase a blu ray reader and hack it. The reason a ready packaged sensor is not available cheaply is simply because of economics and market size.
I'm not convinced by the market/economics argument. Cheap and small ToF sensors exist, even though the market initially was also small.
ToF sensors have huge applications. Initially ToF sensors also cost thousands of dollars. What huge applications do you see that should have driven the prices of over the counter integrable interferometers down? As I stated blu ray (and DVD, CD and laser disc before it) readers are tiny purpose build inferometers. And they are cheap.
Since when are CD reader optics interferometers? The linked video shows a CW homodyne LIDAR used for measuring vibration frequency and counting vibration amplitude in fringes.
Last I looked CD readers used a 4-detector sensor's differential low-pass signals for closed-loop track-following so the rotation need not be optically centered. I also see no reason why optical disc readouts would need homodyne let alone heterodyne readout.
The pits and valleys are spaced λ/4 apart. The reason there is such a stark difference in intensity is because this λ/4 spacing causes interference at pit->valley and valley->pit transition points. Of course this not a standard interferometer but rather a purpose build one. Wikipedia says:
> Interferometry is a technique which uses the interference of superimposed waves to extract information.
And a blu ray player directly uses interference of superimposed waves to extract information. It squarely fits in the homodyne category.
I was always wondering if every car around you had a Lidar, would these systems be confused by light emitted by other cars ?
The first table is already outdated, there are COTS units like the STL-27L that operate on dTOF with a 25m range, return thousands of points per sweep, and fit into a package with a footprint the size of a typical wristwatch. Bulky optics are no longer a requirement, and that's even on the very low end of the price spectrum, more expensive units can do 30+ m in a similarly sized form factor.
Very informative. I assume this is the type of industry work PhDs in physics do