LIDAR (out of main)

[Doggydogworld]

Show me a $10k 360° LIDAR system that isn't going to kill a fast motorcycle driver that (lawfully) overtakes the FSD car from behind in rain while the car is doing an unprotected left turn...
Additionally, most Waymo LIDAR experience is with ~$75,000 class mechanical LIDARs from Velodyne - which have adequate resolution.

Waymo developed their own LIDARs because Velodyne was waaaay too expensive, and a pain in the ass to boot ("the Comcast of LIDAR", haha). How much does Waymo's 360 degree roof LIDAR cost? We don't know. But open up an Ouster unit - it's just a couple of small chips on a spinner. They price against Velodyne because it's dumb to give money away, but their $24k OS2-64 would cost a few grand at 10k unit volume. Either Waymo is near that price themselves or they are complete idiots for continuing inhouse development instead of doing a deal with someone like Ouster.

Solid state LIDAR takes us below $1000 and eventually below $100, but as you note performance is not there yet and the time frame to achieve adequate performance is uncertain. This is an issue for consumer cars, though. Robotaxis can easily afford 5-10k.

To go with the motorcycle example: at night a lawfully driving motorcycle will be spectacularly illuminated by its own headlights,...

​

Unless his light isn't on. Or the motorcycle is a deer. Or a jogger. Heck, cross traffic without headlights is not a corner case. It's very common.

LIDAR systems have to illuminate it with their own source of photons, which have several orders of magnitude lower intensity and are also double-distance attenuated and reflection attenuated by having to travel from LIDAR to the motorcycle, reflect from it exactly towards the LIDAR and back.

Self-driving cars don't use "LIDAR systems". They use LIDAR+Camera+Radar. With proper engineering, that beats Camera+Radar. Period.​

Or as Elon said it: LIDAR is a local maximum that makes it harder to find the absolute maximum.

Local maximum is just buzzword-ism. If radar was $20k and bulky and LIDAR was cheap/tiny, Tesla would use Camera+LIDAR and Musk would talk about radar's useless "single pixel view".

Tesla can't use LIDAR, so they've come up with a clever approach which might obviate the need for it. That's what makes free markets great - people try different stuff and compete for customers. And Tesla's FSD business model is impossible to beat, and Waymo is trying hard to squander their robotaxi lead. This battle will rage for some time. Meanwhile, cheerleading by tribe members who refuse to acknowledge even a single advantage of the competition's approach gets really old.​

 

[mfel]

A technical question, not a pro/con LIDAR but something I've been wondering about:
How do all these systems, LIDAR, radar and ultrasonics, keep from interfering with other vehicles with same?

Simplest example, if two cars with same ultrasonics are side by side, what the heck is happening with their signals so that one car knows which returns are actually from its own pings? Some kind of frequency chirping or gating to overlap?

And what kind of errors show up in LIDAR if every car on road has them? I can understand when there's only a few cars having them, it would be relatively easy to identify and discarded as anomalous a very occasional return. But what about dozens of cars next to each other in traffic?

Which leads to question (putting on blackhat): what about jamming these devices? Take a bright enough laser, or even a right optical band LED "flashlight", and point it at the car, and it is blinded for at least a portion of its view. I guess for LIDAR it is relatively narrow angle which would be blinded, and the same thing could happen to cameras. Though if cameras are filtered to visible light, it would at least be obvious to humans riders what is happening.

 

[ReflexFunds]

Waymo developed their own LIDARs because Velodyne was waaaay too expensive, and a pain in the ass to boot ("the Comcast of LIDAR", haha). How much does Waymo's 360 degree roof LIDAR cost? We don't know. But open up an Ouster unit - it's just a couple of small chips on a spinner. They price against Velodyne because it's dumb to give money away, but their $24k OS2-64 would cost a few grand at 10k unit volume. Either Waymo is near that price themselves or they are complete idiots for continuing inhouse development instead of doing a deal with someone like Ouster.

Solid state LIDAR takes us below $1000 and eventually below $100, but as you note performance is not there yet and the time frame to achieve adequate performance is uncertain. This is an issue for consumer cars, though. Robotaxis can easily afford 5-10k.

I don't see how you can reconcile your views that is its impossible for Tesla to reduce its EV powertrain costs another 50% to match mass market ICE powertrain costs, and yet you think it is extremely easy to bring down high quality lidar costs 99.9% from $75k to $100 in just a few years. LIDAR cost reduction is not trivial. Waymo has made many of the easy changes to reduce their LIDAR cost 90% to $7.5k (but they have 3 of these on each car so $22.5k per car), this will continue to come down quickly, but 100x breakthroughs in technology production cost are never easy, instant or guaranteed.

 

[RDoc]

A technical question, not a pro/con LIDAR but something I've been wondering about:
How do all these systems, LIDAR, radar and ultrasonics, keep from interfering with other vehicles with same?

Simplest example, if two cars with same ultrasonics are side by side, what the heck is happening with their signals so that one car knows which returns are actually from its own pings? Some kind of frequency chirping or gating to overlap?

And what kind of errors show up in LIDAR if every car on road has them? I can understand when there's only a few cars having them, it would be relatively easy to identify and discarded as anomalous a very occasional return. But what about dozens of cars next to each other in traffic?

Which leads to question (putting on blackhat): what about jamming these devices? Take a bright enough laser, or even a right optical band LED "flashlight", and point it at the car, and it is blinded for at least a portion of its view. I guess for LIDAR it is relatively narrow angle which would be blinded, and the same thing could happen to cameras. Though if cameras are filtered to visible light, it would at least be obvious to humans riders what is happening.

With automotive radar, the signal has an id embedded in it so the transmitting unit only responds to the returns from itself. I imagine that lidar and ultrasonics could do the same.
As far as jamming, the acceptance angle is pretty narrow so, particularly for lidar, the jammer would have to be extremely bright. Possible, but this kind of attack is much easier with a human driven vehicle, such as the laser attacks on commercial planes.

 

[MC3OZ]

AM vs FM: The battle brewing in lidar technology | ZDNet

This quote is interesting:-
"According to advocates of doppler lidar, conventional AM lidar is highly vulnerable to interference from sunlight and other sensors. It's also computationally intense and error-prone in the way it deduces the velocity of objects over multiple frames of data. AM lidar uses all kinds of computational tricks to determine the velocity of objects, which is made more complex by the high error rate caused by lighting inconsistencies and sun glare."

If FM lidar is cheaper and simpler then it may shift the balance, it is also newer, I would like an advocate for AM to list the limitations of FM, because I am sure it has some.

 

[ReflexFunds]

AM vs FM: The battle brewing in lidar technology | ZDNet

This quote is interesting:-
"According to advocates of doppler lidar, conventional AM lidar is highly vulnerable to interference from sunlight and other sensors. It's also computationally intense and error-prone in the way it deduces the velocity of objects over multiple frames of data. AM lidar uses all kinds of computational tricks to determine the velocity of objects, which is made more complex by the high error rate caused by lighting inconsistencies and sun glare."

If FM lidar is cheaper and simpler then it may shift the balance, it is also newer, I would like an advocate for AM to list the limitations of FM, because I am sure it has some.

I haven't read much about FM lidar and i'm not a radar expert either, but this just sounds like they are making lidar more like the radar Tesla is already using (just in a different range of wavelengths). I'd guess many of the advantages of FM lidar vs AM lidar would be the same advantages that made Tesla choose FM radar as its backup sensor over AM lidar in the first place.

Tesla is currently using this radar from Continental Continental Automotive. Continental's documents are contradictory on whether this is a pulse compression radar or a FMCW (Frequency Modulated Continuous Wave) radar. Does anyone know if it could be both? (perhaps alternating between modes to get the advantages of both)
Tesla has been developing an in-house radar for some time. We have heard very little about this project and I'd be very interested to know if the project is primarily focussed on cost/vertical integration, or if they plan to significantly upgrade the radar capabilities vs the current Continental device.

 

[MC3OZ]

Karen's opinion in LIDAR:- https://twitter.com/enn_nafnlaus/status/1334298458782658568

So why LIDAR? What does it bring? Expensive? Check. Awkward? Check. Power hungry? Check. Slow refresh rate? Check. Eye/camera hazards? Check. Identification? Worse than cameras. Weather sensitivity? Worse than cameras and way worse than radar.

IMO a FSD solution needs to solve vision regardless. Mostly those using LIDAR are combining it with HD maps specifically because they haven't solved vision.