Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register
  • The final cut of the 9th episode of the Tesla Motors Club Podcast, featuring Chad Schwitters, the former president of Plug In America, is now available. You can watch it now on YouTube or listen to it on all major podcast networks.

Dr. Mary (Missy) Louise Cummings should be opposed for NOT really being expert on FSD for cars

wooter

Nou ik heb niet te klagen over Tesla support
Moderator
May 3, 2017
8,624
8,057
Belgium
The problem is that LIDAR what Audi uses in their adaptive cruise control systems, is of a much lower resolution than the 360º spinning LIDAR setups in Waymo cars.

They're not the same.

So they don't influence each other on volume for mass market use.

Bottom line is that LIDAR does not give context, only distance. Camera don't give context either, but with CV you can create context.

What I'm saying is: this is LIDAR you wish:

1641419799898.png
And based on vague shapes, you can recognise a wall and a car on the left, and something that looks like a sign on the right. And vehicles up ahead. You can also recognise trees, but you cannot read what's on the sign. You need camera's for that.

But if you use a camera, you can recognise all those items too. So what advantage does LIDAR give?

Here's another example:
1641419912878.png


You can recognise a person, but that's about tit For the rest, you have no context where you are.

Velodyne's own system gives this result:

1641420305747.png


Again, you can recognise people, but that's because we make a recognition of the low resolution shape and try to build context. In effect, our mind is reconstructing context based on vision, not on the 3D point clouds.

LIDAR works on the principle that a laser beam will shine infrared light on an object, and the reflection is captured by the sensor. It then determines the distance of the reflecting surface by calculating how long the infrared laser light was travelling, using the speed of light.

Camera's don't bother with sending out infrared light. They just capture existing light and register them on their sensor. They lose distance data, but capture more colours. And with CV, you can recognise objects because you have a wider color gamut.

If you go at the level of front facing LIDAR applications for adaptive cruise control, like Audi's, you are looking at the following resolutions:

1641420238919.png


You could claim that you'd need LIDAR and CV to have a complete picture, except CV and LIDAR have so much different results, so you're just working to find a way to have both stacks recognise the same thing. And what will you do if both the LIDAR and CV stack don't recognise the same thing? Who will you trust?
 
  • Like
Reactions: MontyFloyd
Cellphones and robot vacuums have LIDAR these days. How much did the first digital cameras cost? The first RADAR systems? The price of new tech goes down over time as volume goes up. What part of a LiDAR system do you think is fundamentally expensive?
Volvo and Mercedes halve announced models with LIDAR too.
One can buy a LiDAR module for $27.59, so you would completely trust your life on that?

People here are completely missing the big picture about LiDAR, its cost to benefit is near $0.
Tesla did the work, they concluded it was a waste of $ in a car (but necessary for SpaceX Dragon capsule)

Make a performance matrix, LiDAR has the least value.

Looking at the Xpeng P5 LiDAR location, perfect place to get mud on it, be blinded by any rain and wet roads, and one bad parking spot or pothole could bend the mount.

I do see one area a LiDAR may be useful, but secondary to cameras.
 
Yep. That’s why talking about the cost of automotive LIDAR is somewhat silly until someone actually finds a mass market use for it. How much did radar units cost before they started using them for adaptive cruise control? My guess is that they were not cheap.
It remains to be seen whether LIDAR will enable L3 highway systems but if it does I expect there will be enough volume to drive costs down. Hopefully the Volvo and Mercedes systems will actually get deployed this year.
I think it is already a mass market. For example there are dozens of industrial robots where LiDAR is perfectly acceptable.
 

JHCCAZ

Electrified Engineer
Supporting Member
Feb 2, 2021
537
979
Tucson
I more or less accept the premise that Lidar is not required for basic autonomous driving, even L4 autonomy. I also understand the difficulty (not the impossibility) of camera + Lidar fusion. I think that the development of vision-based pseudo-Lidar approach has probably made such fusion a lot easier, though the usual argument is that it also makes it unnecessary or even confusing.

This may be less of a paradox than it seems, as we get into "helpful redundancy" as an element of a very high reliability system, i.e. the March of Nines topic. Some level of effort in resoving contradictions has to be dealt with if you are to have any kind of backup/watchdog redundancy. To put it another way, it's a valid question to ask "Which one are you going to believe when they disagree?" but I'd also say it's not acceptable to let that objection be the final word and stop there - you actually have to deal with that question and come up with rules or AI inferences to decide. Otherwise you're stuck with no redundancy and that, I believe, puts a cap on the ultimate reliability of the system in edge cases.

I'm very happy to have the existing Tesla Autopilot, but as the L2 responsible driver I also have to be the one to decide whom to believe when it occasionally beeps an urgent collision warning for no apparent reason. And the inverse situation applies also: the still-unfinished software is happy to drive straight into many kinds road barriers or unidentified debris. If I weren't willing to decide how to resolve these conflicts, I couldn't drive the car with AP engaged.

This brings me to my suggestion of how Tesla could best use an inexpensive forward-looking Lidar. The strength of such a sensor would be its ability to detect random debris, ditches and potholes, and whatever other solid objects or pavement defects that may impede the driving path but not be recognizable in the current storehouse of NN-labelable objects. Furthermore, its ability to detect the vertical texture of the road surface would help to reassure the system in the very common cases of difficult shadows, cracks, asphalt-repair confusion and heat mirages. In this latter aspect it would be perhaps not a "alarm watchdog" but more of a calming "seeing-eye dog" consultant. Sandy Munro has been advocating for a FLIR camera; perhaps that's a reasonable alternative with many of the same advantages, a different set of strengths and weaknesses that would still compliment division cameras.

While I think this specific kind of Lidar or similar assistance need not be very expensive at all going forward, I'm pretty sure that Elon / Tesla will be resistant to deploying such a thing. The same way I think they'll be resistant to adding corner cameras that I think would be immensely helpful. That doesn't mean I think they're crazy or stupid, it's just where I come down in the judgment of what could be really helpful despite the incremental costs and engineering effort.
 
  • Informative
Reactions: MontyFloyd
Tesla did the work, they concluded it was a waste of $ in a car
I'm in full agreement that it's a waste of $ in a car unless it allows you to deliver a feature that customers will pay for. Would Tesla be able to charge more than $10k if they added LIDAR? Probably not.
The problem is that LIDAR what Audi uses in their adaptive cruise control systems, is of a much lower resolution than the 360º spinning LIDAR setups in Waymo cars.

They're not the same.
Of course not, it should be no surprise that consumer products are not as good as the state of the art. If you look at the performance of the Valeo 2 unit that Mercedes is using what would it have cost 10 years ago? What does it cost now? I haven't heard any explanation for why costs won't continue to come down.
Camera's don't bother with sending out infrared light. They just capture existing light and register them on their sensor. They lose distance data
Isn't this the only reason to use LIDAR at all? That vision systems can't currently match its performance for depth information.
And what will you do if both the LIDAR and CV stack don't recognise the same thing? Who will you trust?
You trust the one that tells you you're going to run into the back of a firetruck. haha. Isn't Tesla going to have same issue with their voxel NN?
 
  • Like
Reactions: Sporty
This brings me to my suggestion of how Tesla could best use an inexpensive forward-looking Lidar. The strength of such a sensor would be its ability to detect random debris, ditches and potholes, and whatever other solid objects or pavement defects that may impede the driving path but not be recognizable in the current storehouse of NN-labelable objects. Furthermore, its ability to detect the vertical texture of the road surface would help to reassure the system in the very common cases of difficult shadows, cracks, asphalt-repair confusion and heat mirages. In this latter aspect it would be perhaps not a "alarm watchdog" but more of a calming "seeing-eye dog" consultant. Sandy Munro has been advocating for a FLIR camera; perhaps that's a reasonable alternative with many of the same advantages, a different set of strengths and weaknesses that would still compliment division cameras.
That is also what I support, a F-LiDAR as an alternate to RADAR, for accurate range and help with errata ID.
Cost monster still applies.

FLIR is interesting thought. This will certainly ID other vehicles and man-made objects vs natural ones, especially in poor weather.
 

daktari

Active Member
Jan 21, 2017
1,095
1,287
Norway
That is also what I support, a F-LiDAR as an alternate to RADAR, for accurate range and help with errata ID.
Cost monster still applies.

FLIR is interesting thought. This will certainly ID other vehicles and man-made objects vs natural ones, especially in poor weather.
FLIR is not best for cars. Tires, engines and exhaust lights up. EVs only have hot tires. Pedestrians and animals is the use case. Was fantastic in my 2019 etron.
 

Attachments

  • Screenshot_20220106-184612_Photos.jpg
    Screenshot_20220106-184612_Photos.jpg
    162.1 KB · Views: 30

daktari

Active Member
Jan 21, 2017
1,095
1,287
Norway
Following a Tesla?
Yes, that is an X. Electric cars are abundant here in Oslo.

Note the tele-lens Audi uses. It is best for animals and pedestrians at speed. Matrix full beams will flash at objects also (in dark areas).
Night Vision will also draw bounding boxes around these objects. Yellow if no danger, red if anticipating creature crossing into driving path. I would love to have those indicated in a huge AR HUD (see small hud on picture), not only IC.
Note the guy hiding beneath the green light, hard to see IRL but easy for NV.
 

Bladerskb

Senior Software Engineer
Oct 24, 2016
2,687
4,501
Michigan
If cost were no object, I tend to think that sensor fusion between camera and lidar is more reliable than camera and radar. C/L combo gives you high res distance, velocity, and color, but cannot overcome heavy weather. If camera and lidar are rendered useless by weather, the use of a backup radar will not allow the car to drive autonomously because alone, radar does not have the requisite resolution to make any good perception decision.
I would love to point out that it isn't camera vs lidar or even camera and lidar vs. camera and radar.
Its camera, radar and Lidar vs. camera only.

Secondly Radar resolution has gotten better. Waymo for example uses 4D high resolution imaging radar.
In comparison to Tesla's current radar which gives 400 points per second, Waymo and Mobileye's next gen radar give you around 500k points per second.
That's 500k vs 400 in radar resolution. That's over 1250x times more resolution.

Its crazy how people never ask the question... Why is Tesla having a hard time with sensor fusion, could it be that its because they are trying to fuse 1 million pixels from their camera with 40 random radar points from an old obsolete ACC radar? Imagine overlaying 40 random points all over your FOV, how beneficial would it be or how capable would you be to fuse it with what you see?
 

wooter

Nou ik heb niet te klagen over Tesla support
Moderator
May 3, 2017
8,624
8,057
Belgium
Its crazy how people never ask the question... Why is Tesla having a hard time with sensor fusion, could it be that its because they are trying to fuse 1 million pixels from their camera with 40 random radar points from an old obsolete ACC radar? Imagine overlaying 40 random points all over your FOV, how beneficial would it be or how capable would you be to fuse it with what you see?
Got a quote on those 500k LIDARs?
 
Got a quote on those 500k LIDARs?
This is the LIDAR going into the Volvo XC90 EV. Supposedly $500-$1,000 in volume.
It's 300 points per square degree.
120 degress horizontal, 26 degrees vertical, which would multiply up to 936k points but I think it might switch between narrow and wide scan modes so that's not probably not right.
 
  • Like
Reactions: JHCCAZ and Dan D.

Bladerskb

Senior Software Engineer
Oct 24, 2016
2,687
4,501
Michigan
I never quoted what the prices of the Tesla cameras.
I should have sourced the price of radar (my bad), but a quick search shows eBay asks around $200 (you can search more).
Read article a few months ago and I probably am mistaken the cost estimate.
Ebay...you can't be serious right? We are talking about ASP that oems pay in bulk not a random unit someone is selling on ebay.
Again, I was quoting price of system, FROM A TRADE JOURNAL which our company uses in our work.

Links to them, please.

Bottom line, I put an actual quote down from an industry journal with HARD numbers.
Those are not hard numbers. You need to get a new trade journal.
Price of lidar are freely available.
Volvo gets Luminar for $500.
Mobileye gets it for $1000.
Livox sells single unit lidars to any online for $1299 (Horizon).
Its safe to say their mass production version they sell to OEMS in bulk like xpeng is well below that.
Huawei lidars are around 500 with goal to get it under 100-200
I could keep going..Innoviz One is under $1000, Innoviz Two is aiming for 70% cheaper.
Again i could keep going.. Innovusion sells at $1,000 a unit
Which is very rich because your fist line says "this is why people need to get info from direct sources" which is exactly what I did IRT LiDAR.
LOL

no you didn't as i proved above.

Your definition of "plenty" is unusual.
You only list one that is in production and a few others just starting Jan 2022 or coming in months.
As for one in production:
Xpeng P5 came out 2 months ago, 244 sold in September (probably 10K ~ 20K now) but only China and Europe.
12 ultrasonic sensors, 13 cameras, 1x 5mm RADAR and 2x LiDAR (!!!), all for a price between $25,000 to $35,000.

Details on P5 LiDAR is limited, looks like a narrow arc or perhaps a cone of scanning, some closeups here.
Interestingly this article also states LiDAR "Not long ago, a lidar sensor cost about $75,000." but I am sure that talks about systems like the WayMo suite of 4 or more 90* to 360* wide scanning systems.

How Xpeng can profitably sell a car with more than 2x the sensor array of a any Tesla for $10k less is, well, add your adjective here.
Its called google, i wont keep doing your homework for you when it takes a-couple seconds of google search,
Xpeng uses Livox's HAP with horizontal FOV of 120 degrees and The P5 with the two lidars cost only $31k

There are plenty, dozens. i just listed a few. By the end of 2025 there will be 3-5 million high res lidars being sold annually.

In production:
Huawei arcfox (3 lidars)
Xpeng P5 (2 lidars)
Lucid Air

Coming in 2022:
Volvo
NIO ET7
Nio ET5
SAIC
Xpeng G9
Great Wall Motor Saloon EV (4 lidars)
WM Motor M7 (3 lidars)
IM L7
Avatr 11 ( 3 lidars)
etc..etc. there are dozens more
 
So let me get this straight TS you take all this time to write some anonymous internet keyboard warrior hit job on this woman? Bravo you loser.

This f’ing forum and Mods are unreal.

Yeah the TS and other posters saying “good post” are all not biased in anyway for Tesla huh?

I stopped reading about 2-3 sentences in in this post.
 
So let me get this straight TS you take all this time to write some anonymous internet keyboard warrior hit job on this woman? Bravo you loser.

This f’ing forum and Mods are unreal.

Yeah the TS and other posters saying “good post” are all not biased in anyway for Tesla huh?

I stopped reading about 2-3 sentences in in this post.
It's not bias for Tesla on the part of the original poster that's important here. It bias by Mary Cummings against Tesla that's the issue. Her various written statements and other factors would lead the most casual observer to conclude that her anti-Tesla bias is undeniable. This is a poor attribute for a government "expert".
 
Ebay...you can't be serious right? We are talking about ASP that oems pay in bulk not a random unit someone is selling on ebay.

I tried 3 different terms, that is what kept popping up (with hundred pages of various results).
You link is very informative, but the prices are only for chipsets, not complete units. A start.

Its called google, i wont keep doing your homework for you when it takes a-couple seconds of google search,
Knowing the ideal search terms makes all the difference.
You have knowledge and are updated on news, know what is relevant, so adding also those links is immensely helpful for us to read relevant sources and not end up in Google dead ends.

Thanks, sincerely.
 

Products we're discussing on TMC...

About Us

Formed in 2006, Tesla Motors Club (TMC) was the first independent online Tesla community. Today it remains the largest and most dynamic community of Tesla enthusiasts. Learn more.

Do you value your experience at TMC? Consider becoming a Supporting Member of Tesla Motors Club. As a thank you for your contribution, you'll get nearly no ads in the Community and Groups sections. Additional perks are available depending on the level of contribution. Please visit the Account Upgrades page for more details.


SUPPORT TMC
Top