Autonomous Car Progress

[diplomat33]

As near as I can tell from Google maps, Chandler, though an incorporated city, is really just a chunk of greater Phoenix. How useful is it really to have a (robo)taxi service that only serves Chandler (or a portion thereof)? Will Waymo serve all of San Francisco, or just a portion? For a taxi or robotaxi service to be useful it needs to be capable of serving a whole city. Not just the small part with perfect roads and uniformly good signage.

Yes, robotaxi services need to serve an entire city to be useful. And eventually, they will. But the tech is not ready yet for large scale deployment. I am sure Waymo will eventually serve larger areas when the FSD tech is ready. For now, it makes sense to have limited geofenced areas while you are still working on the tech. They give you a decent sample of the driving while keeping your costs at a minimum. Once the tech is good enough, they will offer robotaxis in larger areas.

Why are they opening in S.F. when they still cannot even cover all of Phoenix?

I think because SF offers more interesting edge cases to help "solve FSD" faster. Covering all of Phoenix would not really make sense since it would not really help them "solve FSD faster". It would just give them more of the same of what they are doing in Chandler. It makes more sense for waymo to expand to lots of small geofenced areas in multiple cities than to expand in one big area that offers more of the same. It gives them more diversity in the edge cases as well as more experience with ride-hailing in different city cultures.

I see present robotaxi services as development platforms, not serious public services. They are a step in the right direction, but they are light-years away from competing with taxis and ride-shares, much less getting people to abandon their private cars.

Yes, because we have not "solved FSD" yet. Right now, L4 robotaxis are basically prototypes. Any new tech, in the early days, will be more of a concept. But once the tech matures, then it becomes ubiquitous. Heck, there was a time when the personal computer was seen as an expensive novelty and people though there was no real future for it. But now, everyone has multiple computers. It will be the same with robotaxis. Once the tech matures, robotaxis will be everywhere and will just become a normal part of life.

 

[diplomat33]

Both Snow banks are ever changing. They can get smaller/larger in the space of a few hours. Day time melt then refreeze. You get more snow. The snow bank gets larger. Maybe a lot larger depending on snow fall amount and roads getting plowed making a small snow bank massive in just a few hours
View attachment 717509

My guess is that the HD maps would be created when there is no snow (like in the summer). That would give the car the basic understanding of where the road should be. Then the autonomous vehicles use their sensors like cameras, radar and lidar to detect the snow banks no matter their size, and the car could navigate safely. Autonomous vehicles do not just use HD maps alone to know where to drive.

 

[EVNow]

Yes, because we have not "solved FSD" yet.

Not picking on this particularly - but I think in general we have to recalibrate our expectations.

Its not like FSD will be "solved" on a particular day. Its just a "march of zeros" as Musk said couple of years back. Question is when will FSD be 2x (or 10x) better than humans - in conditions we find in 90% of US roads.

For eg., is Waymo better than humans on Chandler, AZ roads (the part they covered) ? If so by how much. Then, what do they need to increase that area of coverage ? I do feel Waymo missed a golden opportunity to slowly expand the coverage instead of sitting still for 3 years (concentrating on disengagements / 1M miles, I guess). Can't believe a silicon valley tech company allowed that ...

 

[diplomat33]

Not picking on this particularly - but I think in general we have to recalibrate our expectations.

Its not like FSD will be "solved" on a particular day. Its just a "march of zeros" as Musk said couple of years back. Question is when will FSD be 2x (or 10x) better than humans - in conditions we find in 90% of US roads.

I agree. I am using "solve FSD" loosely to mean when autonomous driving is deemed good enough to deploy on a large scale with no safety driver. Eventually, we will achieve enough 9's that we will be able to do that.

For eg., is Waymo better than humans on Chandler, AZ roads (the part they covered) ? If so by how much. Then, what do they need to increase that area of coverage ?

All good questions. I am sure the Waymo engineers and execs are looking at the safety data and making those determinations.

I am looking forward to the 2021 CA DMV disengagement report, to see if there is a notable improvement in total miles and disengagement rate.

 

[Terminator857]

For those asking cost to charge a robotaxi:
.09 cents in Norway krone, $1.04 US dollars, to charge Model 3 from 0% to 100%.

https://twitter.com/x/status/1444774412113285120

 

[Terminator857]

As near as I can tell from Google maps, Chandler, though an incorporated city, is really just a chunk of greater Phoenix. How useful is it really to have a (robo)taxi service that only serves Chandler (or a portion thereof)?

Last I checked Waymo no safety driver didn't even cover all of Chandler.

Will Waymo serve all of San Francisco, or just a portion?

Not SF bay area. I googled for coverage map but came up empty. SF is a relatively small city area wise.

For a taxi or robotaxi service to be useful it needs to be capable of serving a whole city. Not just the small part with perfect roads and uniformly good signage.

Even better needs to serve the metro area.

Why are they opening in S.F. when they still cannot even cover all of Phoenix?

They don't want cruise to make a big splash. In other words they want to show the world who is who in the driverless world.

I see present robotaxi services as development platforms, not serious public services.

Agree.

They are a step in the right direction, but they are light-years away from competing with taxis and ride-shares, much less getting people to abandon their private cars.

Waymo can compete, while losing billions each year.

Even something as simple as being able to leave your dog in the car or make multiple shopping stops without having to take all your purchases from earlier stops into each store will insure that few people will switch away from private cars.

I'm not much of physical shopping person, especially true if you are talking about making multiple stops. Aren't people moving to online shopping?

 

[Terminator857]

... sitting still for 3 years (concentrating on disengagements / 1M miles, I guess). ...

3 years ago Waymo had trouble merging onto a busy freeway. I haven't heard that it is solved yet. It is a difficult problem from multiple angles. Legally someone must let you in, and they can change their mind and hit you and it is the car attempting to merge fault. Bad press. From a non legal perspective you have to play a game of chicken with the cars that you need to merge into. Perhaps a robot taxi can win this using some non conventional technique, like pull out some massive looking armament(s) to win the game of chicken.

 

[Dutchy Ron]

As Waymo gains more and more experience, and solves more and more edge cases, their autonomous driving will work reliably in more and more places. It will take less and less time to expand to the next city.



Unlikely. Even if Tesla solves L4/L5 in the US, there are regulatory hurdles in the EU, not to mention different roads, different traffic rules in both the EU and China. There will be entirely new edge cases. To get L4/L5 working in the EU and China will require more retraining of the NN. It won't happen in one year.



I don't know for sure but the accuracy is probably around 1-10 cm. So they are more accurate than Toyota's 40 cm HD map.

Lexus TeamMate is using HD Maps from TomTom with less than 10cm accuracy…. tested 7 years

 

[daniel]

Not picking on this particularly - but I think in general we have to recalibrate our expectations.

Five years ago I had hopes of being able to buy a driverless car in a decade. Not a certainty, but a real hope. Now I don't expect to own a driverless car ever, given my age. I don't even expect robotaxis where I live in my lifetime, though that's immaterial, because if I have to quit driving, a robotaxi is no better than a ride-share.

I'm just grateful for EAP. And in a few years I may trade up to whatever Tesla has by then. Though I don't expect Tesla, or Waymo, to ever be able to drive safely on South Kihei Road. Heavily-trafficked, two-lane road, no shoulders in places, occasional cyclists and pedestrians who have to use the driving lane (because no shoulders). You need to be able to anticipate that a car approaching you may crowd or even cross the center line to avoid hitting a cyclist or pedestrian on their side of the road. Will FSD be able to do this with anything less than General AI?

 

[EVNow]

Lexus TeamMate is using HD Maps from TomTom with less than 10cm accuracy…. tested 7 years

Here is an article / "review" on TeamMate that reads more like an advertisement. So much for independent media. As they say - without financial freedom you don't really have any freedom.

Both Teammate and Tesla's Autopilot deliver a 360-degree view of the vehicle's surroundings, but one is like having Terminator vision, and the other is the relative equivalent of waving a stick around yourself while blindfolded.

Anyway, it looks like NOA.

 

[diplomat33]

@Bladerskb This should be very interesting. Both Dolgov and Anguelov will be talking about Waymo's AI and ML approach:

https://twitter.com/x/status/1445064021971927041

Looks like we might get an update on Waymo's latest FSD progress.

 

[Dutchy Ron]

Navigation specialist TomTom has equipped more than 3m vehicles with high-definition maps accurate to a few centimetres that get near real-time updates from crowdsourcing. “It’s a continuous stream of data that’s coming in, and that data is exploding — it’s exponential growth,” says Willem Strijbosch, TomTom’s head of autonomous driving.

TomTom’s Strijbosch argues that the Level 4 cars are “overfitted” with technology because they let costs run amok and need to make up for their small sample sizes. But even if ADAS data is less robust, it can be multiplied by a far greater number of real-world hours and eventually close the gap with Level 4 tech. “By the time we’ve driven so many billions of kilometres, in millions of vehicles that have this whole sensor set-up, all corner cases will have been captured,” he says.

See this article in the Financial Time for more information

Robotaxis: have Google and Amazon backed the wrong technology?

An evolutionary approach to building driverless cars is challenging the companies targeting a ‘moonshot’ solution to full automation

www.ft.com

In my opinion HD Maps are required for safety reasons and this will be mandatory for Level 3+ by legislation in the near future.
Last year Volkswagen started Cariad for their software and HD maps. VW will also use MSFT Azure and MSFT Connected Vehicle Platform. MSFT is using TomTom OTA HD Maps as well, just like Stellantis (FCA and PSA), Baidu, Huawei and others.

The National Highway Traffic Safety Administration will come with the results of the investigation in November 2021. My advise for Elon is to use or buy TomTom as soon as possible because it takes 7 years to build a global HD Map. That is the reason why all a start-up map makers are focusing on the local robot taxi, so they only need to map the city one by one.

Qualcomm bids on Veloneer, Nvidia bought DeepMap….. all big tech companies acquire the last AV companies to make everything in-house, just like Elon. Better regret now not using HD Maps than be to late to use them.

I think Tesla has legacy already and needs a new platform for their new cars and future models.

 

[diplomat33]

@Bladerskb Abstract of what Anguelov will discuss in his keynote on Thursday at 12:45pm EST. I am super looking forward to this.

In this keynote, Drago Anguelov, Head of Research at Waymo, discusses Waymo’s progress towards building a scalable technology stack for autonomous driving vehicles. With more than a decade of experience in solving autonomous driving, Waymo is now operating the world’s first commercial ride-hailing service Waymo One in Phoenix and has recently welcomed its first riders in San Francisco by kicking off the Trusted Tester program.

Drago will give an overview of the key autonomous driving challenges and describe how Waymo is leveraging the cutting edge ML systems across the stack to handle them. He will also outline promising avenues to keep expanding the scope of ML in the stack in the future and showcase some of Waymo’s work in the space.

Link to event. It is free:

Scale TransformX Conference | Oct 19th - 21st

Join Scale AI for a three-day, virtual conference as we explore the shift from research to reality within AI and Machine Learning.

scale.com

Original Tweet:

https://twitter.com/x/status/1445426059147677708

 

[diplomat33]

https://twitter.com/x/status/1445782301586051073

On Wednesday, GM announced its next-generation driver-assist system called Ultra Cruise that will enable hands-free driving in most situations when it launches in a Cadillac in 2023.

First announced in 2020, Ultra Cruise is being touted by GM as a "door-to-door hands free driving" solution that will work on all public paved roads in the U.S. and Canada. That equals more than 2 million miles of road, and eventually up to 3.4 million miles.

Ultra Cruise goes beyond Super Cruise. GM says it can acknowledge and react to permanent traffic control devices, provide drivers information via a dynamic display, follow navigation routes, follow posted speed limits, perform both automatic and on-demand lane changes, perform left and right turns, avoid objects, and park in a residential driveway

While capable, the system will require driver engagement. GM touts the system as delivering hands-free driving across 95% of scenarios. Complex intersections and certain situations will still pose an issue. The system, like Super Cruise, will shut down if a driver doesn't take over when the vehicle requests help.

Like Super Cruise, GM notes that Ultra Cruise is a Level 2 driver-assist system, and vehicles equipped with it will not be self-driving cars. Like Super Cruise, vehicles equipped with Ultra Cruise will have a driver-attention monitor that watches the driver's eyes.

It seems that Ultra Cruise will be similar to Tesla's FSD Beta and Mobileye's Super Vision. So, we are seeing several companies now trying to offer L2 "door to door".

I know it is not autonomous driving but it is news about a company's ADAS progress so I figured this thread might still be ok, rather than start a new thread.

 

[rxlawdude]

https://twitter.com/x/status/1445782301586051073

It seems that Ultra Cruise will be similar to Tesla's FSD Beta and Mobileye's Super Vision. So, we are seeing several companies now trying to offer L2 "door to door".

I know it is not autonomous driving but it is news about a company's ADAS progress so I figured this thread might still be ok, rather than start a new thread.

I guess you missed the news that GM's technology is on hold due to chip shortages. And that this new technology won't be available until at least late 2022.

So, vapor against Tesla. Not surprising.

"Ultra Cruise-enabled vehicles will come equipped with lidar, which is rare for production vehicles thanks to the high costs associated with the laser sensor. “The sensing architecture is all new,” Ditman said. “There are additional cameras and radars and we are adding lidar to the vehicle.”

Ultra Cruise won’t be able to handle every driving scenario. Ditman gave the example of a roundabout as a type of complex road condition that the ADAS will not be able to navigate."

FSD handles roundabouts.

Next?

 

[EVNow]

https://twitter.com/x/status/1445782301586051073

It seems that Ultra Cruise will be similar to Tesla's FSD Beta and Mobileye's Super Vision. So, we are seeing several companies now trying to offer L2 "door to door".

I know it is not autonomous driving but it is news about a company's ADAS progress so I figured this thread might still be ok, rather than start a new thread.

Yes - this sounds like FSD Beta. This would be very interesting as a comparison if & when it actually launches.

 

[diplomat33]

I guess you missed the news that GM's technology is on hold due to chip shortages. And that this new technology won't be available until at least late 2022.

So, vapor against Tesla. So you.

No, I am aware of the chip shortage. I am not claiming that Ultra Cruise is coming out now. The article mentions that GM is aiming for 2023 for releasing of Ultra Cruise. I am just sharing the news.

 

[rxlawdude]

No, I am aware of the chip shortage. I am not claiming that Ultra Cruise is coming out now. The article mentions that GM is aiming for 2023 for releasing of Ultra Cruise. I am just sharing the news.

Sharing news of a product that does not exist, and will demonstrably be more expensive than FSD due to the addition of Lidar. And "mapped roads."

As Spock would say, "how crude."

 

[EVNow]

No, I am aware of the chip shortage.

That chip shortage is an interesting case - we'll have to see how it gets resolved. Chip makers say they won't fund facilities that use old tech that most car companies use.

I'm actually surprised chip shortage hit Ultra Cruise - isn't MobilEye part of Intel ? Doesn't Intel have a few foundries ...

 

[Bitdepth]

That chip shortage is an interesting case - we'll have to see how it gets resolved. Chip makers say they won't fund facilities that use old tech that most car companies use.

I'm actually surprised chip shortage hit Ultra Cruise - isn't MobilEye part of Intel ? Doesn't Intel have a few foundries ...

I think the fact that the compute platform is on 5nm node rules out Intel as a viable foundry, which leaves TSMC or Samsung as the only two options.