First L3 Self Driving Car - Audi A8 world premieres in Barcelona

[lux_cars]

Those above are what is happening on the zFAS that is soon shipping with the Audi A8. Here is a bit about what to expect of the next zFAS version. The Audi Highway Pilot (60 to 130 kph driving with automated lane changes) and the Audi AI Zone (autonomous parking in a multi-story):

Audi Elaine concept car

It also details some other interesting things Audi is working on regarding driver comfort, monitoring and assistance.

I think this is a pretty comprehensive picture of what to expect from Audi's driver assistants up to 2019 - as far as the "competition watch" goes. Many of these will probably appear on the Audi e-tron quattro, similar to Audi A8, I'd guess in its second model year, if not the first...

"The new generation processor has more computing power and refined sensors with even greater range and precision. The Audi Elaine can therefore serve its owner as a highway pilot, relieving the burden on the driver over extended distances. The highway pilot is an extension of the traffic jam pilot in the A8 and allows piloted driving at speeds from 60 to 130 km/h (37.3 to 80.8 mph). When the highway pilot is active, the Audi Elaine changes lanes automatically – it can pass and then return to the original lane, for example. It can initiate, perform and complete such actions autonomously, without any involvement of the person behind the wheel. If the driver does want to intervene, they can do so spontaneously at any time.

The networking of the car and the infrastructure is crucial for driverless mobility. The driver parks the Audi in a designated area (the handover zone) and exits the vehicle. From there, the car drives automatically and unoccupied into a multistory parking garage offering a variety of services, such as a car wash, a package station, a gas station or a charging post. Thanks to Audi AI, the car does all of this itself.

Connected with its surroundings, the intelligent Audi can even locate an unmarked parking space on the side of the road and pull precisely into it. At the desired time, the vehicle is back in the handover zone, ready for its next journey. Drivers can follow the actions of their vehicle at all times and even add new tasks using an app.

Customers of the premium brand will soon be able to save time and enjoy greater convenience in Audi AI Zones. Audi is currently developing a standard interface for a wide variety of smart devices and is preparing web-based, vehicle-specific apps. The project is almost ready for production. The Audi Elaine will thus become an IoT device (internet of things), seamlessly and smartly integrating itself within the world of its user."

 

[Snuffysasa]

Actually I already pointed it out long ago that it wasn't launching this year. For the journalists who attended the launch event, they suggested that it wasn't launching yet until 2019.



That goes back to the point I made with @Bladerskb about how until they actually sell vehicles equipped with the option, the claims of being first don't really apply (and why I don't value development news as highly as a metric because of stuff like this where the timelines are misinterpreted or sometimes changed).


@AnxietyRanger, I would also like to point out that Audi did not go around and correct the misinterpretation of Audi's timeline by the media. You have previously suggested that Tesla had a obligation to do so, but in this case Audi didn't do so either. As I said, I don't see manufacturers doing this.

That's a mistake. The feature will be launching in the 2019 A8 in the USA. The 2019 A8 will be launching in the US in the later half of 2018 with L3 Traffic Jam Pilot enabled.

2019/2020 Audi will be launching the full speed version, Highway Pilot.

I'm really disappointed in Audi. That they didn't release partial of their software as ADAS. instead the Audi A8 only has ACC and stupid lane departure warning.

They could have easily had the best single lane full speed highway adas.

However I'm proud of Nissan, they might be the first automaker (might even beat Tesla and Audi) to release software based on their SDC resources.

Their L2/L3 system in 2018 will be based on their SDC software and utilize eyeQ4 (8 cameras, multiple radars and lidars) and will use a complete REM map of japan highways built by mobileye and Zenrin

I am very excited for this nissan vehicle as well! I knew it would have a complete map of Japan with Zenrin and Mobileye and I assumed eyeQ4. But I did not know 8 cameras and Lidar!! How do you know this??

Also isn't this vehicle going to be level 2 now.. Why say L2/L3?

 

[AnxietyRanger]

That's a mistake. The feature will be launching in the 2019 A8 in the USA. The 2019 A8 will be launching in the US in the later half of 2018 with L3 Traffic Jam Pilot enabled.

You have a source for this?

Anyway, to explain @Snuffysasa's comment timeline, for anyone who has not followed Audi launches, usually it goes a bit like this - certainly for the past two Audi A8s for example:

- New model launches in late year X (let's say X = 2017 for example) and goes on sale in Germany first
- It goes on sale as the "X+1 model" (e.g. "2018 model" in this case)
- Over the next few months it goes on sale in Europe as well
- Then around week 22-23 in year X+1 (e.g. 2018), the model becomes the "X+2 model" (e.g. "2019 model"), sales start that same spring
- The X+2 model usually gets some features that did not make the initial launch ramp-up (e.g. Traffic Jam Pilot possibly)
- It is the X+2 model that goes to the U.S. as well, sales starting when the stock arrives mostly...

I can't think of any model where the X+1 model-year would have gone to the U.S. It is the X+2 due to homologation but also due to the fact that U.S. is an inventory market and they need to build and ship inventory there...

I personally have no info whether U.S. will get Traffic Jam Pilot in the X+2 model, but I think X+2 model is when some markets will.

 

[lux_cars]

2019/2020 Audi will be launching the full speed version, Highway Pilot.

What’s the source for this? Full-speed highway pilot would be fantastic!!

 

[Bladerskb]

I am very excited for this nissan vehicle as well! I knew it would have a complete map of Japan with Zenrin and Mobileye and I assumed eyeQ4. But I did not know 8 cameras and Lidar!! How do you know this??

Also isn't this vehicle going to be level 2 now.. Why say L2/L3?

Nissan initially were aiming for L3 but it looks like they will settle for L2. It doesnt reslly matter as long as it works. This could be the first consumer product based on fsd r&d software

 

[AnxietyRanger]

So that Nissan system has 5 radars, 4 lidars and 8 cameras in that video. That is very nice 360 degree redundant coverage.

 

[Bladerskb]

So that Nissan system has 5 radars, 4 lidars and 8 cameras in that video. That is very nice 360 degree redundant coverage.

yes its very interesting.

With that i will say that while ppl like @stopcrazypp think i don't give my share of fair criticism to other companies. I have said multiple times in the past that Ford and Toyota are clueless when it comes to SDC and will be the very last.

I have also always believed that if Audi doesn't activate their L3 traffic jam pilot in 2017, it will be absolutely useless in 2018/2019.
Why? because in late 2018 you will see more advanced L2 cars coming on the scene like the current gm supercruise which has taken the lead and the coming nissan propilot 2.0 and possibly even tesla EAP.

We don't know if GM will use Cruise tech in their next generation super cruise 2.0 but i would assume so.

A L3 under 37 mph becomes unattractive in 2018. This sorta displays the slowness of trad automaker and their inability to pivot and change.
While the announcement and Audi's vision was great in late 2014 when there were no good full speed L2 systems. They should have pivoted when they saw the incumbents were introducing good high speed L2 systems. They could have easily had the best L2 system by far period and did what Nissan was planning on doing!

But no they stuck to their guns like blackberry and probably upper management are now regretting their decision as they see GM supercruise flourishing in media coverage and will be even more dismayed if the sales of the feature supercruise pushes GM sales forward and increases their profit margins.

While they are stuck with a system they can't activate for 12+ months maybe?

This is why its good to be Agile. GM saw that they couldn't differentiate their supreme system from the poisoned well of l2 systems so they knew they had to do something about the 15 seconds timer. They delayed their launch one last time and added the camera facing sensor which was a last minute addition and it allowed them to introduce and claim the "First Hands Free L2 system".

50% of automakers will go bankrupt in the next decade because they refuse to change just like the blackberries and Nokia of the last decade.

Unfortunately for Tesla their real challengers are just as silicon valley as they are; "GM cruise" and "Google waymo" and while Mobileye was hampered by the automakers of the past, their unprecedented lead in computer vision have saved them and kept them as front-runners.

Unfortunately there are two automakers who refuse to be disrupted and are willing to pivot and adapt. Its GM and them allowing cruise to run solo speaks volume and then you have Nissan who has consolidated their 5-10 ADAS team and their next system in 2018 will be based off their SDC team and software with 100+ employees.

 

[AnxietyRanger]

Fair points @Bladerskb. Interesting to see how it plays out. Basically you are saying Audi has been so focused on autonomous, they've neglected the Level 2 driver's aids and their rise in the meanwhile. Fair.

Audi's reluctance to introduce a full-speed Level 2 steering system indeed seems a product of the "traffic jam assit to traffic jam pilot to highway pilot" strategy, which is a meticulous way of avoiding mode confusion because nothing "Level 2" will ever go above their "Level 3"...

As you've said before, there is logic to Audi's approach, but it is possible - well, very likely - their efforts will be seriously eclipsed (at least in the short term) in the *driver's aid* space if they won't pick up pace. That is an unfortunate byproduct of their strategy.

 

[jimmy_d]

I thought only in the US they had the reputation, not their own home country. I guess they are internationally known like that.

They're probably not alone there. Lots of announcements getting made (including by Tesla).

 

[jimmy_d]

Audi Ankündigungsweltmeister

but no Fahrvergnügen?

 

[jimmy_d]

Nissan initially were aiming for L3 but it looks like they will settle for L2. It doesnt reslly matter as long as it works. This could be the first consumer product based on fsd r&d software

Nice!

Was excited, and surprised to hear on the video that propilot was available starting in 2016. Apparently only in Japan, so I went to look at it on Nissan's site and see if I could find some user evaluations by people who had owned it.

So the Japanese version being sold so far is just a single forward camera - no radar or sonar, and the Nissan brochure and a few japanese news articles that I read describe the tech as a combination of Mobileye and Nissan home grown stuff (Nissan claims their contribution makes it safer in one brochure, whatever that means). Generally it seems to offer capabilities that don't differ from what I imagine a pure mobileye implementation would offer. I'm impressed that a single-camera only system is out there in the wild and it generally gets OK reviews from drivers, it seems. Looks like it's mainly for use on highways, which is where the manufacturer recommends it. But it also looks like users are trying it outside those restrictions as well, so I guess it must not be geofenced.

The press coverage is pretty softball stuff (which Japanese press is, generally), but it's possible to find customer evaluations on car review sites to get more balanced opinions.

Here's an article on a PR site: 高速道路で自動運転！日産「プロパイロット」は想像以上にすごかった

I read through a half dozen user reviews and people seem to generally find it convenient once you get used to it's quirks but say you need to keep an eye on it. Some users recommend particular places that you might be tempted to use it but probably shouldn't. One user was annoyed that it becomes unusable (ceases to function) whenever the road markings are less than great.

If you're interested: 価格.com - 『プロパイロットの弱点』 日産 セレナ 2016年モデル のクチコミ掲示板

Here's a translation of the opening comment for the above user discussion thread:

---------------
I drove it for 1000 kilometers after making a delivery. I would use it whenever I could on regular roads, regardless of speed. Although I think it's very convenient it is just a machine. There were a few things that did not work as I expected, and occasionally it does something scary. I think it's possible to avoid scary stuff as long as you stay alert so that you can intervene at any time.

Honestly I think I'd be asking for an accident someday if I got overconfident.

Since I think that the chance of accident will fall if you know where the system is weak, I hope to share the weaknesses of propilot.

The strange behavior I felt was:

· It's vulnerable to hills and declines. Maybe the road surface to leading car contact point is used to judge distance. For instance when the car is going downhill and the car ahead is already climbing the next hill the system seems to misjudge the distance and tries to close the gap to the leading car.
· It tends to speed up going downhill and on a steep downhill it won't slow down even if it exceeds the set speed. When a car ahead was waiting for a signal downhill from me I couldn't trust the car to stop on it's own and had to brake myself.
· There were times when it didn't see bicycles and motorcycles. I get the feeling it can't see narrow things.
· When it was cloudy or there were shadows on the road it would lose sight of a greyish, asphalt colored car ahead of me and suddenly accelerate, then suddenly brake just before a collision. There was what seemed like an emergency alarm and sudden braking. That seemed unreliable.
· It tended to lose sight of lanes when in a traffic jam. I think that maybe lane marking is being blocked by the car ahead. It comes back it as soon as the front car moves.

Have other people using the system experienced this?
--------------

You can google translate other parts if you're curious (above was translated by me) but the gist is the same. This thread is mostly people sharing scary stories, but other places where people give an overall evaluation they seem to think it's great as long as you're careful with it.

 

[Bladerskb]

Nice!

Was excited, and surprised to hear on the video that propilot was available starting in 2016. Apparently only in Japan, so I went to look at it on Nissan's site and see if I could find some user evaluations by people who had owned it.

So the Japanese version being sold so far is just a single forward camera - no radar or sonar, and the Nissan brochure and a few japanese news articles that I read describe the tech as a combination of Mobileye and Nissan home grown stuff (Nissan claims their contribution makes it safer in one brochure, whatever that means). Generally it seems to offer capabilities that don't differ from what I imagine a pure mobileye implementation would offer.

whats the point of this post? everyone know that pro-pilot 1.0 sucks and it sucks because it was developed by a small team of about 5-10 people. like most ADAS system. it shows that Nissan barely worked on this.

How is it that you cant see the difference of a software developed by hundreds of people like AP or propilot 2.0 and a software developed by 5?

"pure mobileye implementation?" lmao take that crap someone else.

If you think that the limit of mobileye eyeq3 is pro-pilot 1 then keep your BS over at the H2.5 thread. I left that thread because of you ppl's buffoonery. Don't mess up this thread too.

 

[S4WRXTTCS]

Nissan initially were aiming for L3 but it looks like they will settle for L2. It doesnt reslly matter as long as it works. This could be the first consumer product based on fsd r&d software

From a drivers standpoint the difference between L2 and L3 are monumental as you're fully aware of.

Now it might not matter software/technology wise because it might be more of a regulatory issue, or some decision made outside of a purely engineering one. Something that isn't a reflection on whether it works or not.

But, if it's made L2 instead of L3 due to engineering reason then it doesn't really work.

I'm a little concerned that a lot of cars will start to stack up on the advanced L2, but not allowed to be L3 line. Or such a low speed for L3 that it's not entirely useful.

Can't wait for someone to bust through that line to a car capable of L3 on a rainy PNW evening at 80mph.

 

[S4WRXTTCS]

whats the point of this post? everyone know that pro-pilot 1.0 sucks and it sucks because it was developed by a small team of about 5-10 people. like most ADAS system. it shows that Nissan barely worked on this.

How is it that you cant see the difference of a software developed by hundreds of people like AP or propilot 2.0 and a software developed by 5?

"pure mobileye implementation?" lmao take that crap someone else.

If you think that the limit of mobileye eyeq3 is pro-pilot 1 then keep your BS over at the H2.5 thread. I left that thread because of you ppl's buffoonery. Don't mess up this thread too.

You're scolding him for talking about a video YOU posted?

Come on dude.

No, everyone does NOT know about the history of ProPilot.

He also made it clear that he was only talking about ProPilot 1.0, and we know from the video you posted that it's not a reflection on the next version of it.

His post was useful information for those of us that are curious about the existing implementations, and future implementations of autonomous or semiautonomous cars. I didn't feel as his post was anything more than that.

Once again your attitude and ego are putting people off. To the point where people aren't even going to bother posting anything informative out of fear of your wrath.

I also don't get what point there is in leaving a thread. The people are the same so what difference does it make? If you like talking about autonomous cars then by all means stick around, but try to be a decent human being.

 

[stopcrazypp]

You're scolding him for talking about a video YOU posted?

Come on dude.

No, everyone does NOT know about the history of ProPilot.

He also made it clear that he was only talking about ProPilot 1.0, and we know from the video you posted that it's not a reflection on the next version of it.

His post was useful information for those of us that are curious about the existing implementations, and future implementations. I didn't feel as his post was anything more than that.

Once again your attitude and ego are putting people off. To the point where people aren't even going to bother posting anything informative out of fear of your wrath.

I also don't get what point there is in leaving a thread. The people are the same so what difference does it make? If you like talking about autonomous cars then by all means stick around, but try to be a decent human being.

I had the same feeling. I didn't know about the quality of Propilot and I found @jimmy_d's post very informative. And given we have very few members here that can read Japanese and that Propilot 1.0 had only sold to consumers in Japan so far (won't be coming here until the 2018 Leaf which won't be available until January next year), I'm pretty sure close to no people here have read about consumer impressions about the system from daily use.

@jimmy_d the translation and information is appreciated.

 

[AnxietyRanger]

whats the point of this post? everyone know that pro-pilot 1.0 sucks and it sucks because it was developed by a small team of about 5-10 people. like most ADAS system. it shows that Nissan barely worked on this.

I have no information regarding any intent, but IMO the way it comes across, the point of that post becomes sounding "objective", yet just happen to quote a massive post about scary stuff and leave a mention of great stuff to a single small sentence at the end. It is hard to think that these threads are nothing but driven by biases, when this happens.

And I'm not sure I could avoid it either. When you take a position, you tend to cement your's in whatever you do - and others posting contrary views easily adds motivation to keep doing more and more of that. We feed off of each other and this is what goes around, comes around.

Look at that post regarding Pro Pilot 1.0. How could you not write that very same post about AP1 if you wanted too, yet that would not happen on the other side of the debate? Of course you could. And of course one side wouldn't. You can find scary threads for AP1. So the post kind of tells nothing, because in the end it makes no attempt at being a look at the whole of it, but for whatever reason ends up seeking and elevating critique of it...

Frankly, as much as I wanted to press Informative on jimmy_d's post, I just can't because of the obvious bias. I'm actually more inclined to believe you simply saying it "sucks", because you at least have formulated an (informed?) opinion on the system and are not trying to sugar-coat it into anything else. But I won't press Informative on you either, because there is no info there, just an opinion (well, there is unverified info on team size).

Not to even mention that the real point of that video is the ProPilot of the future, but for some reason it all became about the ProPilot that is currently out there. This comes back to my point about giving benefit of the doubt. Would the same person discuss Tesla's AP2 in the same manner as they did Nissan's ProPilot efforts? Of course not...

 

[Bladerskb]

From a drivers standpoint the difference between L2 and L3 are monumental as you're fully aware of.

Absolutely.

Now it might not matter software/technology wise because it might be more of a regulatory issue, or some decision made outside of a purely engineering one. Something that isn't a reflection on whether it works or not.

But, if it's made L2 instead of L3 due to engineering reason then it doesn't really work.

Not Quite. You have to remember that the every SDC prototype no matter how far along they are, they are more advanced than the best ADAS out there. Why? Is it the load of sensors on them or the powerful computers in the trunk? Actually not.

Its simply the vast amount of control algorithm that has been programmed for the car.
Almost all automaker use the same tech in their ADAS as in their self driving prototype.
That tech is the mobileye eyeq3. The mobileye chip and cameras are the primary sensors of almost all automakers and their state of the art algorithms utilize every info from that chip.

Audi's A7 Jack L4 Prototype for example uses one main FOV camera with eyeq3. The radars are the exact same radar on all ADAS cars.
The front lidar doesn't actually make a difference because this is not the 360 degree lidar that can map out the environment but a very limited amount of lasers for object detection use.

So the primary information you use to drive the car is the same camera and eyeq3 that exist on all ADAS.

The difference being the software, the control algorithms. The reason today we don't have L2 cars as advanced as these SDC prototypes is because automakers are not agile and their teams are split. You have a program with a team with hundreds of employees and dozens of partners working on SDC in one location and then you have another team of a-couple employees working on your ADAS.

As a software engineer, I won't be surprised if someone told me that a single employee wrote the control algorithms for most companies ADAS.

Whether its Audi who drove 550 miles or Delphi who did a cross country completely and fully autonomous, or Volvo with their Drive Me L4 Car or Nissan and their pro-pilot prototype, or Honda, BMW and more. They all use Eyeq3 as their primary sensor. All these companies have one thing in common, they don't use 360 laser. So they depend on the accuracy of Eyeq3 painting them a accurate picture of the world and its objects.

There are several reasons you will make a car L2 instead of L4/L3. One is regulation, another is lack of developed redundancy in sensor/systems, another is rate of disengagement. If you were aiming for failure rate of 1 in 300,000 miles and you are at 1 in 100,000 miles. You can simply make the car L2.

I'm a little concerned that a lot of cars will start to stack up on the advanced L2, but not allowed to be L3 line. Or such a low speed for L3 that it's not entirely useful.

Can't wait for someone to bust through that line to a car capable of L3 on a rainy PNW evening at 80mph.

I don't think you will see a L3 car, no other automaker is gunning for L3 anymore. Not even audi who planned to transition from L3 traffic jam pilot to L3 highway pilot. You might eventually see audi l3 traffic jam pilot fused with their L4 ambition and become no more.

secondly automakers don't even have a desire to improve their ADAS. all the money and resources in billions are being funneled into their SDC program. The actual internal ADAS team (if you can even call it a team) sees none of it. I bet the engineers are itching to make updates but can't because the business refuse.

 

[jimmy_d]

@jimmy_d the translation and information is appreciated.

Thanks for the positive feedback. Originally I figured people could just google translate the page but then I tried it on a section an it was pretty much illegible, so I thought I'd translate one comment to give a sense of it. Not a great translation and just one comment, but glad to hear that someone thought it was interesting.

I had previously looked into what people were saying about the volvo system here:

Auto Pilot Assist 2 thread (experience, problems)

The commonality of experience between those folks and Tesla folks, and now the pro-pilot folks makes for an interesting commentary on the state of ADAS system IMHO.

 

[S4WRXTTCS]

I don't think you will see a L3 car, no other automaker is gunning for L3 anymore. Not even audi who planned to transition from L3 traffic jam pilot to L3 highway pilot. You might eventually see audi l3 traffic jam pilot fused with their L4 ambition and become no more.

secondly automakers don't even have a desire to improve their ADAS. all the money and resources in billions are being funneled into their SDC program. The actual internal ADAS team (if you can even call it a team) sees none of it. I bet the engineers are itching to make updates but can't because the business refuse.

I can certainly understand your excitement about an SDC based system even if it is only L2. I'm excited as well purely from an engineering standpoint because like you I see that as finally being on the right path. It's certainly closer to the end goal.

At the same time you're going to have a really hard time selling me an L2 system for two fundamental reasons.

The first one is my experience with an L2 system has shown me that I lose too much situational awareness when it's driving. So to me an L2 system is a cheat where the Automaker can put the liability on me while at the same time having a car that drives itself. This is one of the reasons I don't use AP much. I don't because it's good enough that I find myself seduced by it like a beautiful girl smiling at me, but then the second I allow myself to be drawn in the boyfriend shows up and rudely interrupts the moment.

The second one is it doesn't really give me much over what I already have. The Tesla AP1 might be a primitive L2 system, but it gives me what I expect out of a L2 system.

I don't really expect the regulatory hurdles to be all that much in the United States. It certainly isn't shaping up to be that way. Instead I think it's more about public acceptance, liability, and slowly white listing roads. Redundancy is certainly a hurdle, but I think that's a solvable problem.

I can't say I'm going to be disappointed if there isn't an L3 system. My concern was we'd see all kinds of L2 systems, but no one willing to transfer liability from the human to the machine.

 

[AnxietyRanger]

Interesting thoughts there @Bladerskb, thanks for sharing.