Autonomous Car Progress

[diplomat33]

That said the only really interesting thing in there to me was just how bad Fords system still is... It was well over a year ago Sandy Munroe did a video riding in a Ford Blues Clues highway ADAS car and noted how it was incapable of handling anything more than the mildest curves on highways and would disengage often due to highway curves.... and it appears they still haven't done anything to address that.... an issue even GM was doing well on years prior and Tesla has had largely solved even longer.

Ford disbanded Argo and took some of their engineers to work on Blue Cruise. They probably did that because they know the system needs a lot of improvement. So maybe Ford can adapt Argo's autonomous driving stack to improve Blue Cruise? Although Ford was working with Argo for a few years. It's odd they did not think to ask Argo to help with with Blue Cruise sooner. It is telling that Blue Cruise is still so bad.

 

[nativewolf]

Ford was far behind everything EV 8 years ago. It's been a giant push for them to even get 2 EVs out there. They did though, not bad vehicles either. I think Ford was expecting the collaborative projects on self driving to have had more success. GM to its credit has Cruise at least putting something into the field even if it seems a bit rushed and lacks optimization. In this regard GM is doing fairly well.

 

[kabin]

CNBC’s Michael Wayland recently tested ADAS from Tesla, GM and Ford:

I drove hundreds of miles 'hands-free' in GM, Ford and Tesla cars – here's how it went

CNBC's Michael Wayland recently tested "hands-free" systems from Tesla, GM and Ford. Theirs are among the most readily available and dynamic on the market.

www.cnbc.com

Overall, I think it is pretty fair. I don't agree that it is hard to know if FSD is on or not. The blue wheel icon on the screen is easy to see in your peripheral vision so you don't need to take your eyes off the road. Also, there is a very characteristic chime sound when the system turns on or off. So it is not an issue for me.

Good stuff. Glad it wasn't a hit-job so common from major media outlets. I have to give TSLA credit for taking chances. Good to hear GM is doing relatively well.

 

[Bladerskb]

Thanks. Good insights. You mention a few times that the issues are likely caused by Cruise driving in SF. It sounds like you do think a lot of the issues are caused by Cruise overfitting to SF. Basically, Cruise has done most of their driving in SF so their autonomous driving is overfit for SF. The good news, as you point out, is that as Cruise does expand to other places like Phoenix and Austin, they will be able to retrain their autonomous driving to be more generalized.

Yup and it looks like it’s driving in downtown Austin is better and of-course the roads are more similar to SF.

https://twitter.com/x/status/1605759041006977025

Second vid:

Google Photos

photos.google.com

 

[Dewg]

Yup and it looks like it’s driving in downtown Austin is better and of-course the roads are more similar to SF.

https://twitter.com/x/status/1605759041006977025

Second vid:

Google Photos

photos.google.com

The pull over for pickup was pretty bad. I thought it would just pull over completely in that gap, close to the curb, but it stopped at an angle with its butt in the traffic lane, and then moved forward back into the lane before the client got in. I wonder why.

 

[diplomat33]

The pull over for pickup was pretty bad. I thought it would just pull over completely in that gap, close to the curb, but it stopped at an angle with its butt in the traffic lane, and then moved forward back into the lane before the client got in. I wonder why.

My guess is the car realized that the pull-over was bad and tried to adjust it. Yeah the whole pull-over was bad. I can see why the SFMTA is mad at Cruise when you got 100+ driverless cars doing this kind of bad pull-over all over the city. Frankly, I am still not super impressed with this video. It's at night when there is almost nobody around it so the Cruise does not have to deal with traffic or pedestrians or even any complicated turns.

 

[Bladerskb]

I mean, the guy you quoted doesn't work for Waymo, so you again appear to be spamming the forum with non-sequiturs.

Waymo, as a reminder, began testing in SF in 2009


To be fair to Waymo though- the post of mine you quoted pointed out they began mapping in LA in 2019... and it appears they want to offer service there in 2023.

So that's only FOUR years, not FIVE between mapping and offering service. QUITE AN IMPROVEMENT!

Assuming they actually deliver on time of course.

So instead of admitting that you were wrong about how long it takes to map a city. You are now insinuating that Waymo is 4 or 5 years behind Cruise. Lmao.

Not only did Cruise prove that HD mapping a city is easy and quick by doing it in mere weeks.

Pony.AI went from 6 months to map and launch in a new city in 2017 to 2 weeks to map and launch in a new city in 2021.
Again the myth that HD map and lidar cannot scale has been debunked at every level.

Anyone still repeating it is knowingly spreading misinformation.

But ofcourse KS can’t ever admit he’s wrong although there’s no shame in someone admitting they were wrong about something.

 

[Knightshade]

So instead of admitting that you were wrong about how long it takes to map a city. You are now insinuating that Waymo is 4 or 5 years behind Cruise. Lmao.

Literally nothing in this sentence is accurate.

Not that your posts were ever anything but terrible, but they've gotten markedly worse lately-- you ok?

 

[diplomat33]

Good news for Waymo.

https://twitter.com/x/status/1606026743005040640

 

[diplomat33]

Dolgov says 5th Gen was a game changer and teases "even more ambitious" plans for Waymo in 2023.

https://twitter.com/x/status/1606053447584792576

Some key improvements in 5th Gen:

https://twitter.com/x/status/1606054667229593600

And 5th Gen is generalized across multiple cities and scaled in SF and Phoenix at the same time and already showing incredible performance in LA from the get-go.

https://twitter.com/x/status/1606055081358413824

 

[diplomat33]

Big update from Baidu:

https://twitter.com/x/status/1607377644542390278

 

[EVNow]

Pony.AI went from 6 months to map and launch in a new city in 2017 to 2 weeks to map and launch in a new city in 2021.

How many cities do we have Pony.AI operating in now ... as in robotaxis open to public ?

 

[EVNow]

https://twitter.com/x/status/1607513131034767361

 

[EVNow]

https://twitter.com/x/status/1608329918882123781

 

[diplomat33]

https://twitter.com/x/status/1608329918882123781

This article is lazy journalism. It takes a few barebone facts, just enough to make their word quota. Yes it is technically true that Cruise is the first to do driverless in 3 cities but the article also leaves a lot out. It is pretty obvious whoever wrote this piece did not do much research.

1) The article leaves out that Cruise is not 24/7 while Waymo is. The article also leaves out that Waymo also has 24/7 driverless in SF, not just Phoenix, and working to launch 24/7 driverless in LA. The article also leaves out the fact that the Cruise services areas in Phoenix and Austin are very small.

2) The last sentence is also completely false since Cruise and Waymo also use a sensor and software-defined platform that improves thanks to machine learning algorithms evolving with an increase in data with the computational power of supercomputers. Waymo and Cruise also use camera vision and use a lot of data and ML to solve vision problems. Furthermore, big data and ML is critical to solving autonomous driving regardless of what sensors you use. In fact, cameras, lidar and radar all use ML. You need ML for everything, from perception, prediction to planning. Cruise and Waymo use big data and supercomputers to train their ML algorithms just as much as Tesla does, if not more.

3) In fact, all AVs are "sensor and software defined platforms" since they all rely on sensors and software to make driving decisions. And all major AV players rely on ML and supercomputer training for all or most of their stack. So making it sound like a "sensor and software defined platform" that relies on ML and training with a supercomputer is what sets Tesla apart is false and ridiculous. Yes, there are differences, notably that Tesla is vision-only and Cruise and Waymo are not. And there are big differences in the software as well. But using big data and supercomputers to train ML algorithms is not the difference between Tesla and Cruise/Waymo.

 

[EVNow]

Yes it is technically true that Cruise is the first to do driverless in 3 cities but the article also leaves a lot out. It is pretty obvious whoever wrote this piece did not do much research.

More importantly ...

The Cruise service is not open to the public in Austin. It is currently restricted to friends and family of Cruise employees, but they are taking actually paid rides.​

 

[DanCar]

Big update from Baidu:

https://twitter.com/x/status/1607377644542390278

Insignificant update if you live in the U.S.?

 

[diplomat33]

Insignificant update if you live in the U.S.?

No, it does not affect the US. But it is significant progress for Baidu and it is significant progress of AVs in general.

 

[diplomat33]

New Cruise driverless video:

0:00 Start
0:09 Pull out for new destination
1:05 No right on red
1:49 Commentary on neighborhood routing
3:12 Flashing red light
3:52 Trying to figure out why the route is weird
4:50 Honking at parallel parking car
5:42 Slowing for unknown (crosswalk?)
6:23 Slowing for unknown (crosswalk?)
8:04 Slowing for unknown
8:49 Unprotected left, no oncoming
9:33 Right lane change due to lane split
10:24 Entering Golden Gate Park
10:44 Driving through a small puddle
12:49 15 mph for unknown reason
15:42 Entering Lincoln Blvd
15:57 Braking for unknown
16:13 Slowing for unknown (crosswalk?)
16:56 Pull over

 

[scottf200]

There's also audio indiications when AP is engaged or disengaged as well... one of several things that they found annoying that if you owned/drove the car for a while you'd realize aren't actual issues.

That said the only really interesting thing in there to me was just how bad Fords system still is... It was well over a year ago Sandy Munroe did a video riding in a Ford Blues Clues highway ADAS car and noted how it was incapable of handling anything more than the mildest curves on highways and would disengage often due to highway curves.... and it appears they still haven't done anything to address that.... an issue even GM was doing well on years prior and Tesla has had largely solved even longer.

Ford disbanded Argo and took some of their engineers to work on Blue Cruise. They probably did that because they know the system needs a lot of improvement. So maybe Ford can adapt Argo's autonomous driving stack to improve Blue Cruise? Although Ford was working with Argo for a few years. It's odd they did not think to ask Argo to help with with Blue Cruise sooner. It is telling that Blue Cruise is still so bad.

It appears that Blue Cruise now handles curves fairly well but goes from a hands-free state to a hands-on/holding state (ala Tesla's AP) on some curves.

Lot of that explanation in this recent video.