Autonomous Car Progress

[Zhelko Dimic]

That's like only watching Fox News and to a lesser extent CNN and then one day coming out to the real world and realizing most of what they were saying is complete BS and not even reflective of reality at all.

You gotta understand that Dave and James Douma are not there to educate you, they are there to brain-wash you. Dave has like what 99% of his network in Tesla? James 85%. These people has literally every bit of their money in Tesla. They are motivated to pump it up as much as they can.

As I said before, i could record a full video clip of all the downright blantly false statements that James Douma makes and I'm not talking about opinion based statements. I'm talking about statement equivalent to saying "the earth is flat" in the AV industry.

For example take a look at the very last Dave interview with James:

Dave: "In 5 years I don't know whose gonna have 60,000 vehicles testing L4 on beta testers, i don't whats their hope, what's their plan"
James Douma : "You have to have 1 million cars with cameras out with people driving around collecting data. I will believe that other people are serious when Ford or GM or VW are shipping cars with boatload of sensors, like an FSD compatible suite, that they don't have software for...When I see other companies doing that then I'll think they are super serious. Why isn't everybody else doing this. I'm confused by everybody else. Why isn't ONE GUY doing this."

What you see displayed is the level of utter ignorance and cluelessness Dave and James has when it comes to the AV industry. First James says you need 1 million. Remember before it was that you need 100k, then 250k, then 500k.. Notice the trend here? Its whatever number of cars that Tesla has/ Its not that you need 1 million cars, its that you need to be Tesla. These are tesla fanatics whose brain can't process anything outside of the Tesla universe. If you don't have the Tesla sticker. Then you're just not cutting it.

Then he says there's not a single company that are shipping cars with a boatload of sensors like whats on Tesla. Not only is he wrong, but they are shipping cars with better hardware and with software already developed for it.

So even if you use James nonsense criteria. His own criteria proves him wrong.

Remember that Tesla AP2 originally ship with 10 TOPs and now with compute of 144 TOPs. Its also worth pointing out that In the world of gathering data, even 1 TOPs is sufficient

Tesla - 12 low resolution (all 1.2 mp) cameras, 12x ultrasonics, 1 forward radar / also removed, 10 TOPS / 144 TOPS

Just to name a few:

1. 2021 Zeekr 001 - 11 (all 8mp) high resolution cameras, 1x radar, 12x ultrasonics, 30-32 TOPs
2. 2021 BAIC arcfox - 12 high resolution cameras, 6x radars, 12x ultrasonics, 3x high resolution lidars, 800 TOPS
3. 2020 Xpeng P7 - 13 cameras 5x 5th gen radars, 12x ultrasonics, 30 TOPS.
4. 2021 Xpeng P5 - 13 high resolution cameras, 5x 5th gen radars, 2x high resolution lidars, 12x ultrasonics, 30 TOPS
5. 2022 Xpeng G9 - 13 high resolution cameras (forward is 8mp & sides are 2.9 mp) , 5x 5th gen radars, 2x high resolution lidars, 12x ultrasonics, 508 TOPS
6. 2022 Volvo - 360 camera, 5x radar, 12x ultrasonics, 254 TOPS
7. 2022 NIO ET7 - 11 high resolution cameras (7x 8mp, 4x 5mp), 5x 5th gen radars, 1x high resolution long range lidar, 12x ultrasonics, 1016 TOPS
8. 2022 NIO ET5 - 11 high resolution cameras (7x 8mp, 4x 5mp), 5x 5th gen radars, 1x high resolution long range lidar, 12x ultrasonics, 1016 TOPS
9. 2023 Cadillac Lyriq - 360 cameras, 5x radars, 1x forward lidar, 12 ultrasonics, 300 TOPS
10. 2022 WM Motor M7 - 11 high resolution cameras (7x 8mp, 4x 5mp), 5x radars, 3x high resolution lidar, 12x ultrasonics, 1016 TOPS
11. 2022 Zhiji IM Motors - 15 high resolution cameras, 5x radars,12x ultrasonics, 508+ TOPS

I could keep going...its pointless.

Now onto the software. Now what does FSD Beta do? Door to door L2 driving. Are there softwares out there in development and about to be developed that will do the same thing and even exceed FSD beta in reliability and performance? YES! James and Dave claim there isn't.

• 1) Mobileye Supervision on the Zeekr 001, OTA in a few months (door to door, works anywhere in china)
• 2) Huawei Autopilot on the BAIC Arcfox Alpha S HI (door to door, works on dozens of mapped china cities, with dozens being added each quarter)
• 3) Xpeng Xpilot 3.5 on the Xpeng P5 (door to door, works on mapped roads, no details on which places are mapped)
• 4) GM Ultracruise on the Cadillac Lyriq (door to door, works anywhere in the US)

So you see that these people are spreading huge misinformation and when you look at their other comments, for example
James Douma basically said other companies have been paralyzed and made "no progress in 5 years". He equates actually having L4 robotaxi running in an city as no progress. But a L2 system that you have to grab the wheel from, to avoid it killing you every acouple minutes as major progress. You realize that these people have absolutely no clue what they are talking about or what's going on in the industry.

There's alot more in that interview that's blantly wrong/misinformation.

I have disagreed with your post, as I feel you attacked Dave and James while attributing them wrongful motives. They're both interested in truth, and don't need to care about money, or pump the stock - @DaveT motivation is to do something meaningful and good for the world, and this is why he keeps accumulating capital trough TSLA holding - it's not to become rich beyond any measurement. After you've been on this board for years, I'm disappointed that you know so little about Dave.

James Douma himself has mentioned in one of the interviews (I'm paraphrasing) that he doesn't track China and doesn't know what is going on there, and it's possible best Tesla competition will come out of there.
Which is in line with your post.

I also found your post informative - where do you get all the info on progress in China?

 

[Terminator857]

... They're both interested in truth, ...

Dave is a super fan like many of the people he interviews. Yes lots/most/all? of people are interested in the truth, but arrive at much different opinions. This isn't so much about truth but conjecture about the future. The truth is that Elon and the super fan base have been wrong time and time again about FSD and will continue to be disappointed.

 

[Zhelko Dimic]

Dave is a super fan like many of the people he interviews. Yes lots/most/all? of people are interested in the truth, but arrive at much different opinions. This isn't so much about truth but conjecture about the future. The truth is that Elon and the super fan base have been wrong time and time again about FSD and will continue to be disappointed.

That may or may not be true, but OP said Dave does this to pump the stock... That was highly offensive to me, knowing Dave a bit

 

[Zhelko Dimic]

Dave is a super fan like many of the people he interviews. Yes lots/most/all? of people are interested in the truth, but arrive at much different opinions. This isn't so much about truth but conjecture about the future. The truth is that Elon and the super fan base have been wrong time and time again about FSD and will continue to be disappointed.

And FSD reflects Elon's chaos Engineering style: changes of leaders, 4 re-writes so far, quick iteration and stubbornness that flips on the dime the moment it's clear given approach won't work... That's why they're late, they kept hitting local maximums, and needing to redo approaches.
I'd never work for Elon. But I'm 50+. It's prob. the best place in the world for young Engineer to cut its teeth.
And Elon will get it done - his persistence and ability to change his mind are real powerful tools.
I project 12-18 months until it's genuinely usable, better than some human in most situations, worse in others. L4 - who knows...

 

[S4WRXTTCS]

Again, your math is flat out wrong here.

YOU paid 5k for EAP, long since 100% recognized as revenue.

And 3k for FSD, which for pre 3/19 remains largely undelivered but let's use YOUR number and say they recognized 1k of the 3k with stop light recognition.


That means they got $6000 in revenue.

I would recommend not doing multiple quotes from different individuals in a response as it confuses things.

Here is what I said:

"The old money pales in comparison to the new money coming in. The only thing left of the $8K (EAP+FSD) that I gave Tesla that's left to being recognized is probably $1K to $2K."

That means the $5K for EAP has already been recognized which is exactly what you say. That only leaves some fraction of the $3K left to be recognized.

The <$3K is what I was referring to as old money.

The new money is $10K/$12K for FSD, and FSD Subscriptions. The FSD Marketing impact also generates a lot of sales.

The point I was making was there wasn't much old money left to be recognized so there isn't much motivation in that. To me the math on that is pretty solid even if I can't quantity what impact FSD marketing has on Sales. But, we don't even have to include that as Tesla sells so many vehicles that even if the take rate is only 10% that's a lot of new money where a lot of it is immediately recognized.

I'm not sure when Tesla will recognize the old money still left to be recognized as there was never a promise of delivery of really anything with pre 3/19 FSD.

Can they recognize it when its still a drivers assist L2?

It might be infinity before they can recognize the remaining old FSD money.

So in a way they made infinity in interest off the money I gave them. That's an impressive kickstarter.

 

[stopcrazypp]

I would recommend not doing multiple quotes from different individuals in a response as it confuses things.

Here is what I said:

"The old money pales in comparison to the new money coming in. The only thing left of the $8K (EAP+FSD) that I gave Tesla that's left to being recognized is probably $1K to $2K."

That means the $5K for EAP has already been recognized which is exactly what you say. That only leaves some fraction of the $3K left to be recognized.

The <$3K is what I was referring to as old money.

The new money is $10K/$12K for FSD, and FSD Subscriptions. The FSD Marketing impact also generates a lot of sales.

The point I was making was there wasn't much old money left to be recognized so there isn't much motivation in that. To me the math on that is pretty solid even if I can't quantity what impact FSD marketing has on Sales. But, we don't even have to include that as Tesla sells so many vehicles that even if the take rate is only 10% that's a lot of new money where a lot of it is immediately recognized.

I'm not sure when Tesla will recognize the old money still left to be recognized as there was never a promise of delivery of really anything with pre 3/19 FSD.

Can they recognize it when its still a drivers assist L2?

It might be infinity before they can recognize the remaining old FSD money.

So in a way they made infinity in interest off the money I gave them. That's an impressive kickstarter.

The point is the "new money" is deferred also in the same amount, due to "City Streets" still not delivered. So it's not like Tesla is pocketing $10k for the new cars and only waiting on the $3k (minus whatever for red lights) of the old ones.

Instead the math for the new cars is they are pocketing the EAP portion (which they have every right to) and deferring the "City Streets" portion. This is made even more clear in markets that are still selling EAP separately (I know this forum is US centric, but let's not forget a world still exists outside the US).

"New money" and old is the same in this regard. Where they differ is after "City Streets" becomes a general release, the money can be pocketed for the new cars, but Tesla may still need to defer for the old ones (due to more extensive promises).

 

[Terminator857]

...
I project 12-18 months until it's genuinely usable, better than some human in most situations, worse in others. L4 - who knows...

Everyone who has gone along with Elon's optimistic exceptions has been wrong. Not much wrong with that, except when you take people's money and don't deliver. In this case the optimists get more done than the realists or the dubious.

 

[Sporty]

Yes, its not suprising that they are starting out with area that is not as busy.

But what is suprising is ... they have been testing for a looooong time already and its taking forever for them to scale within even a single city - let alone top 50/100 cities. Remind sme of all those fantastic battery tech that never scale.

Geoscaling is the Achilles heel of HD Map Robotaxis.

I think that’s left to be seen. I’ve worked on some number of hard problems and there’s a lot to be said for getting to a known working spot and then scaling from that. You have to be systematic about it. I worry that Telsa is not being systematic. You need to dial in success & performance with tests to make sure future improvements don’t cause regressions. They either don’t have that, or they dial those qa tests down or off for some internal reason at times. I personally think they don’t have such tests or not effective ones and depending to heavily on humans to test it. If that’s the case it will never work well. There are too many things to check and too many edge cases.

 

[Knightshade]

Here is what I said:

"The old money pales in comparison to the new money coming in. The only thing left of the $8K (EAP+FSD) that I gave Tesla that's left to being recognized is probably $1K to $2K."

That means the $5K for EAP has already been recognized which is exactly what you say. That only leaves some fraction of the $3K left to be recognized.

The <$3K is what I was referring to as old money.

Yes, and you suggested they've only recognized perhaps 1/3rd of the old money.

Which would mean they have $6000 net recognized revenue from the "old money" people.

We seem to agree.

Where your math goes off the rails is here:

The new money is $10K/$12K for FSD, and FSD Subscriptions. The FSD Marketing impact also generates a lot of sales.

And they only recognize 50% of that based on the last info we have from Tesla.

Meaning they only get.... the same or less revenue from the "new" money compared to the old.

5-6k.


So no- the "new" money is not much more than the old. It's the same or less. (same for 12k buyers, less for 5k buyers).

The point I was making was there wasn't much old money left to be recognized so there isn't much motivation in that. To me the math on that is pretty solid even if I can't quantity what impact FSD marketing has on Sales.

This seems to be a different argument.

They don't get any, of either, until they actually deliver the remaining promised features of course.

But, we don't even have to include that as Tesla sells so many vehicles that even if the take rate is only 10% that's a lot of new money where a lot of it is immediately recognized.

Again this math doesn't really work out.

If the higher price means a lower take rate---as I'm guessing you and Troy seem to think- and the recognized amount is the same old versus new until they deliver more features, they're taking in LESS revenue per X number of sales than when they had a higher take rate.... especially if the EAP-only take rate was significantly higher than the new FSD take rate.

I'm not sure when Tesla will recognize the old money still left to be recognized as there was never a promise of delivery of really anything with pre 3/19 FSD.

Of course there was.

The promised deliverable is easily recognizable as minimum level 4.

Because it promised to be able to conduct both long and short trips with no action required from a person in the drivers seat.

Anything below L3 requires action from a human driver.... Monitoring, being prepared to take over, etc.

Can they recognize it when its still a drivers assist L2?

Nope.

It might be infinity before they can recognize the remaining old FSD money.

So in a way they made infinity in interest off the money I gave them. That's an impressive kickstarter.

Naah.... there'll come a time, though we're probably still a couple years away from it- where if they haven't delivered for the earliest buyers you could file a successful lawsuit for a refund, with interest.

There's at least one pretty long thread that beats this topic to death.... but it largely comes down to what a reasonable person/jury would judge is the useful life of the vehicle.

Some folks in the thread think it's 3 years because they change cars like shirts.

Some think it's the average length of time a person owns a new car- which is like 8.5 years average.

Some think it's longer- but I suspect the 8.5 guys are nearest to having a case (NOT LEGAL ADVICE).


So Tesla is on a clock, just nobody can say for sure when it expires

 

[S4WRXTTCS]

The point is the "new money" is deferred also in the same amount, due to "City Streets" still not delivered. So it's not like Tesla is pocketing $10k for the new cars and only waiting on the $3k (minus whatever for red lights) of the old ones.

Instead the math for the new cars is they are pocketing the EAP portion (which they have every right to) and deferring the "City Streets" portion. This is made even more clear in markets that are still selling EAP separately (I know this forum is US centric, but let's not forget a world still exists outside the US).

"New money" and old is the same in this regard. Where they differ is after "City Streets" becomes a general release, the money can be pocketed for the new cars, but Tesla may still need to defer for the old ones (due to more extensive promises).

I don't think new money and old money can be considered the same amount due to Tesla raising the price of FSD multiple times while not giving Old FSD people much of anything.

Here is the FSD pricing History.

Pre-Revamp -> $3K (most people paid around $3K and some managed to get it for $2K).

Post-Revamp -> On all these I added $2K to account for the Basic AP part of EAP being moved to the base cost of the car.

April 2019	$7,000
May 2019	$8,000
August 2019	$9,000
July 2020	$10,000
October 2020	$12,000
January 2022	$14,000

In April of 2019 it was $7K to get equivalent of EAP+FSD (which I paid $8K for in 2018). So in April of 2019 the price of the FSD portion was $2625. So 37.5% of it.

That percentage likely went down as the price went up due to EAP features like Smart Summon being added, and Tesla raising the price.

The only FSD feature released was Traffic Light and Stop sign control in April 2020. But, it didn't seem to be tied to any price increase.

I would love to see the Percentage of FSD money that Tesla recognizes on each purchase. I think its safe to assume its a LOT because they likely over value the EAP stuff. They also can recognize 100% of an FSD Subscription, but I doubt they have many subscribers right now.

Pricing Source ->

Tesla to raise Full Self-Driving price to $12,000

On January 17th, Tesla is raising its Full Self-Driving price by $2,000, bringing it to $12,000 in the US. What's interesting about this price increase is

www.notateslaapp.com

 

[S4WRXTTCS]

Because it promised to be able to conduct both long and short trips with no action required from a person in the drivers seat.

I agree that was what was being advertised with it being L4. But, they gave themselved a lot of wiggle room. So at best its a soft promise.

The infinity part was a joke, but I think Tesla will stretch it on for so long that people will forget entirely about it.

I expect to get a refund in 50 years because whoever is the CEO of Tesla at the time just wants to get it off the books.

 

[OncomingStorm]

I think that’s left to be seen. I’ve worked on some number of hard problems and there’s a lot to be said for getting to a known working spot and then scaling from that. You have to be systematic about it. I worry that Telsa is not being systematic. You need to dial in success & performance with tests to make sure future improvements don’t cause regressions. They either don’t have that, or they dial those qa tests down or off for some internal reason at times. I personally think they don’t have such tests or not effective ones and depending to heavily on humans to test it. If that’s the case it will never work well. There are too many things to check and too many edge cases.

thinking tesla does no/little QA is a bit of a reach, since the AI can fully take over the car and do whatever it wants i would imagen most changes are probably pretty heavily tested both in simulations and real world. I also would argue they are being quite systematic, its just a vastly different beast then a geofenced robo taxi system, they can't focus entirely on a single area so to scale it up without having to release it in select areas first so they have to take a far different approach.

what i find amusing is that everyone gripes at tesla for their methods because yeah its not perfect and has some issues here and there (tho i'd argue its mostly expectation mitigation that needs to happen) but the way they are building their system it will (and does) out perform any contender i've seen imho.

i would also add that they clearly know about the issues that are commonly complained about, its unreasonable to assume they are ignorant reguardless of how they operate and the fact that they haven't stopped everything to fix it tells me that these are very much rare instances, have very little actual impact on customer satisfaction or are so fundamental to the operation of the AI that they cannot be fixed until more progress is made. in any of those cases all you can really do is sit back and enjoy the over engineered and underregulated features they let us play with by calling them beta haha

 

[stopcrazypp]

I think that’s left to be seen. I’ve worked on some number of hard problems and there’s a lot to be said for getting to a known working spot and then scaling from that. You have to be systematic about it. I worry that Telsa is not being systematic. You need to dial in success & performance with tests to make sure future improvements don’t cause regressions. They either don’t have that, or they dial those qa tests down or off for some internal reason at times. I personally think they don’t have such tests or not effective ones and depending to heavily on humans to test it. If that’s the case it will never work well. There are too many things to check and too many edge cases.

They definitely do regression tests. See 27:10 here where Karpathy talks about validation:

The problem is those tests may not be enough to to cover all the cases that people are seeing. That's why there seems to be so much regression and it's a few steps forward, few steps back.

 

[stopcrazypp]

I don't think new money and old money can be considered the same amount due to Tesla raising the price of FSD multiple times while not giving Old FSD people much of anything.

Here is the FSD pricing History.

Pre-Revamp -> $3K (most people paid around $3K and some managed to get it for $2K).

Post-Revamp -> On all these I added $2K to account for the Basic AP part of EAP being moved to the base cost of the car.

April 2019	$7,000
May 2019	$8,000
August 2019	$9,000
July 2020	$10,000
October 2020	$12,000
January 2022	$14,000

In April of 2019 it was $7K to get equivalent of EAP+FSD (which I paid $8K for in 2018). So in April of 2019 the price of the FSD portion was $2625. So 37.5% of it.

That percentage likely went down as the price went up due to EAP features like Smart Summon being added, and Tesla raising the price.

The only FSD feature released was Traffic Light and Stop sign control in April 2020. But, it didn't seem to be tied to any price increase.

I would love to see the Percentage of FSD money that Tesla recognizes on each purchase. I think its safe to assume its a LOT because they likely over value the EAP stuff. They also can recognize 100% of an FSD Subscription, but I doubt they have many subscribers right now.

Pricing Source ->

Tesla to raise Full Self-Driving price to $12,000

On January 17th, Tesla is raising its Full Self-Driving price by $2,000, bringing it to $12,000 in the US. What's interesting about this price increase is

www.notateslaapp.com

I wasn't arguing the numbers were *exactly* the same for every car (obviously they are not, given even before the price increase, the price you paid varied whether you bought with the car or after, and if there was a promotion running like those that got FSD for $2k etc,), only that they are around the same and being treated largely the same.

But if you want to do numbers analysis, you don't have to really guess, the UK still sells EAP separately: £3,400 (~$4,600) for EAP, £6,800 (~$9,250) for FSD.
Design Your Model 3 | Tesla
As a sanity check, Germany EAP 3.800 € (~$4360), FSD 7.500 € (~$8600), I presume it's roughly the same for rest of EU.
Konfigurieren Sie Ihr Model 3 | Tesla
China: EAP ¥ 32,000 (~$5000), FSD ¥ 64,000 (~$10,000)
设计您的 Model 3 | Tesla
That should cover most of the major non-US markets.

So basically translating to US currency terms, they are counting EAP for $5000, FSD alone for $5000, of which only traffic light/stop sign can be counted as revenue for FSD. So they aren't really overvaluing EAP, it's the same price as it was before (yes, I know about the AP reshuffling and unbundling, but talking about option price). And they are leaving a lot on the table for FSD (perhaps even more than when it was $2-4k alone, depending on circumstances you bought it).

 

[EVNow]

I think that’s left to be seen. I’ve worked on some number of hard problems and there’s a lot to be said for getting to a known working spot and then scaling from that.

Well, I've spent 3 decades as a s/w dev, designer and architect I can tell you if your design is not geoscalable, you have to restart from scratch. We had to throw away many applications that we wanted to take from one country to 100+ countries and restart.

You need to take into account as many complexities as possible when you are designing. But you implement a subset and build on that. Geoscaling is one of those basic features that they should have taken into account in the beginning - if they didn't, thats reflects poorly on them. I don't think they would have completely ignored it - may be they underestimated the scaling issues ? Its difficult to tell ...

You have to be systematic about it. I worry that Telsa is not being systematic. You need to dial in success & performance with tests to make sure future improvements don’t cause regressions. They either don’t have that, or they dial those qa tests down or off for some internal reason at times. I personally think they don’t have such tests or not effective ones and depending to heavily on humans to test it. If that’s the case it will never work well. There are too many things to check and too many edge cases.

Tesla does all that - simulation, internal testing etc. But with NN, many times what is progress for some is regression for others (see the above - this is going to be the issue for Google as well. May be their solution is overfit to small geos they have been testing in).

 

[rbt123]

Geoscaling is the Achilles heel of HD Map Robotaxis.

I'm not sure why that would be.

If software can obtain sufficient knowledge about the area to drive through it in real-time, then processing the same data offline should lead to the same amount, or more, knowledge.

The big advantage to HD map which is impossible in real-time (only) is you know immediately whether an object is static or not, even if you do not know what the object is.

 

[EVNow]

I'm not sure why that would be.

If software can obtain sufficient knowledge about the area to drive through it in real-time, then processing the same data offline should lead to the same amount, or more, knowledge.

The big advantage to HD map which is impossible in real-time (only) is you know immediately whether an object is static or not, even if you do not know what the object is.

This would be true in a static world. Otherwise every new plant would be marked as a dynamic object by Waymo

In a world where keeping 2D maps updated is difficult, how is anyone going to keep HD Maps updated ?

HD Maps is the equivalent of people talking about PV roads that charge cars as they drive through (yes, they were there, arguing thats the only way to make EVs "practical").

 

[AKAlex]

I'm not sure why that would be.

If software can obtain sufficient knowledge about the area to drive through it in real-time, then processing the same data offline should lead to the same amount, or more, knowledge.

The big advantage to HD map which is impossible in real-time (only) is you know immediately whether an object is static or not, even if you do not know what the object is.

I think the big question here is how much manpower went into making that area work. Was it 10 guys for a year? Or 1000?

Tesla's approach does appear as if it will have an advantage in scaling.

 

[diplomat33]

This would be true in a static world. Otherwise every new plant would be marked as a dynamic object by Waymo

In a world where keeping 2D maps updated is difficult, how is anyone going to keep HD Maps updated ?

HD Maps is the equivalent of people talking about PV roads that charge cars as they drive through (yes, they were there, arguing thats the only way to make EVs "practical").

This has been explained so many times. Creating and updating HD maps is not difficult or costly anymore. And HD maps do not prevent scaling.

 

[EVNow]

This has been explained so many times. Creating and updating HD maps is not difficult or costly anymore. And HD maps do not prevent scaling.

Stop it. We are talking about millions of miles of roads worldwide. Not just in a few blocks of some cities. Repeating a lie doesn't make it true.

Since you don't believe Tesla, lets see what others are saying. Both GM & Mobileye say HD Maps don't scale. Why don't you accept the obvious ?

GM relied on lidar-scanned high-definition maps for Super Cruise, but Ditman said it wasn’t practical to map all 2 million miles of road for Ultra Cruise. “We do rely on similar map data,” he said. “However, we have a larger number of sensors that also observe the roads so when we combine the map accuracy with what our sensors see of the road geometry and the road markings, we’re still able to accurately place ourselves and drive the right nominal path.”