Waymo One launches

[Tam]

On the occasion of Waymo One's announcement, PBS did a report on Autonomous Trucks:

Steve Nadig, Daimler's Head Engineer for Mechantronics is not willing to guess that it can replace human driver in 5 or even 10 years.

That despite of its world's first road-licensed self-driving truck in Nevada 3 years ago.

The reporter continued to say the engineering consensus for autonomous trucks is 10, 20, maybe 30 years which will be good for easy long stretch highways but it will still need human for local roads.

What's going on? I thought autonomous technology is either here now or soon but now PBS drops a 30 year time bomb?

 

[strangecosmos]

I’m posting videos of Waymo One as I find them here: Waymo One launches

 

[strangecosmos]

The reporter continued to say the engineering consensus for autonomous trucks is 10, 20, maybe 30 years will be good for easy long stretch highways but still need human for local roads.

What's going on? I thought autonomous technology is either here now or soon but now PBS drops a 30 year time bomb?

There is no consensus and nobody can predict the future. Experts disagree on when full autonomy will happen.

 

[CarlK]

You seem to know little about the car industry. I could fill this page with successful collaboration examples.

Sure you could but that only exposes the weakness of both those companies and the industry. Technology companies don't have that kind of thinking. They don't want to collaborate and share with others. To be the leader is the only game they could play. Autonomous driving is about technology not about car building, which is probably what you alluded to. I recall you once claimed those auto companies are Silicon Valley company too. They are far from it. Sorry but they are hopelessly behind in tech imo.

 

[electronblue]

components that break a lot ... Just bring the van into the garage. ... Different requirements versus the Tesla model.

I really take it that you don’t owa a Tesla?

It doesn’t seem like the sort of thing that would scale well to mass production vehicles, but I do love those cute lil wipers. Waymo’s pure robotaxi model means it can have expensive components that break a lot and everything will be fine. Just bring the van into the garage. If a car gets stranded, you can send out a driver to pick it up and take it back home in manual mode. Different requirements versus the Tesla model.

In all seriousness though. First of all Waymo does not have a ”pure robotaxi model”. What they have is a sensor suite generation they are validating on robotaxis and secondarily on trucks. This is part of a roadmap that will head onto other things over time.

Secondarily all autonomous driving providers will have to solve sensor cleaning in some manner. Indeed it is one of the strongest arguments against Autopilot 2 that it can never be Level 5 capable hardware as Tesla originally announced in 2016 because it has insufficient sensor cleaning. The forward radar is disabled by mere slush and the rearward camera is susceptible to being covered frequently. Aside from the three front cameras all sensors are only ”cleaned” by heat and in the case of the radar covering bumper not even that — and let’s not even go to the ”protective cocoon” of the ”sonars”...

And before anyone says Tesla plans to go vision only riddle me this: Autopilot 2 cameras can not see around the nose. How will a Level 5 self-driving car manuever in parking lots when it can not see this area if the non-camera sensors (and perhaps also the rear camera) are disabled by weather? Level 5. No steering wheel required... Tesla said summon your car from the other coast...

What some of you seem to be missing is the thesis that sensor suite and hardware miniatyrization is a trivial task compared to building the autonomous driving software. Once Waymo has validated their system with larger hardware components slapped onto a car integrating those components and improving on their mechanics if needed could well be expected to be a fairly straightforward task. That is what car companies and car component makers do for a living. But unlike others Tesla has already committed to a suite with their Level 5 capable claims. Failure in this area for Tesla looks very different compared to the others even though they too (and I expect will) can change the suite for future cars.

Building self-cleaning sensors and light sources is nothing new in the industry except for Tesla for whom it seems to be exceedingly hard to integrate such things. These are some of common production components over the past decades:

- Automotive cameras with dedicated washers (eg Audi)
- Automotive headlights with mechanically extending washing system (many brands)
- Automotive headlights with wipers (eg Volvo)
- And of course there are tons of mechanical features like electrically folding roofs etc that work reliably for years and years

These have all been fairly commonly found features in regular production cars.

And we have already seen the entry of automotive cameras with integrated lens spinners and washers for example.

None of this means Waymo will necessarily be the first to consumer car market. Indeed even I don’t expect that but I don’t consider the hardware question the showstopper at all in this. Prices will go down and components will continue to become smaller. Waymo has chosen to validate their technology through fleet development so that is what they will do in first iterations. Once this is done the technology can go to other places as well.

I expect the first car-responsible driving solutions consumers can buy and use to be from MobilEye together with a European premium brand or in Japan. These companies have different roadmaps together with MobilEye to bring levels 3 and 4 to consumers.

In the meanwhile the Version 9 update killed even the automated windshield wipers in my Level 5 capable Tesla.

 

[Bladerskb]

@Tam Mercedes, Toyota and Uber are among companies with the worse autonomous tech so i'm not surprised.

 

[J1mbo]

There is a lidar hidden on this photo.

Forward-facing only though. Am struggling to get the point of that.

 

[electronblue]

Forward-facing only though. Am struggling to get the point of that.

All very logical though since that single forward Lidar is part of Audi’s Level 3 Traffic-jam Pilot which is car-responsible driving on a single motorway lane in traffic so being absolutely sure about same lane, forward objects is sufficient. It also has three forward facing radars and two rear radars to add to it and camera so there is redundancy as well.

Next stage Level 4 piloted motorway driving from Audi uses two forward Lidars they have already shown that prototype. It is still divided road only but faster speeds. But many companies have shown sidefacing Lidars too without cupolas...

 

[J1mbo]

It also has three forward facing radars and two rear radars to add to it and camera so there is redundancy as well.

You have succinctly summed up exactly why I don't see the point.

 

[Bladerskb]

Of course it is. How is someone running a fleet of hundreds or thousands cars each equipped with an expensive Lidar in small limited areas is going to achieve general autonomous driving? There is little hope, if at all, that LIDAR will be cheap and practical enough to put in every car in the forseeable future. No to mention anything they developed with LIDAR, as Elon describes as on crutches, will need to be redone when they switch to other types of sensors. Waymo is not that dumb. Autonomous taxi or ride hailing is exactly the market it's after.

As for the planned Jaguar car purchase Waymo is just to use them to expand its ride hailing or test fleet. It's not to make autonomous car Jaguar or anyone else could sell to customers to put in the garage.



Someone is "working on" something is not the same as someone, borrowing the phrase Elon likes to use, has a clear path to get there. Remember only a couple years ago Mobileye was still dead set against full autonomous driving and was discouranging others to do so. It was even lobbying agencies to slow down the approval process largely because it pretty much owned the driver assistance market and did not want anything to disrupt it. They have to jump on the wagon now but what they had really is not a workable to way to get there.

Whereas Google started its development as one of the moonshot projects without any practicle purposes in mind and selected LIDAR, Mobileye has never had the vision (excuse the pun) of autonomous driving when developing its image chips. It is now trying to use the same technology with the wishful thinking that adding improvements will enable them to get there. They will get better driver assistance but autonomous driving? I don't believe so.

It's pretty clear Tesla saw the "clear path" from the very begining so it chose to put cameras and computers in every car it sells regardless if you purchased the option or not. Not saying there will not be a lot of challenges along the way, or if it will need better brains (processor and learning algorithm), but Tesla is the only company that has set up the ground work to reach general autonomous driving. No one else does. It's pretty clear to Elon when he made that comment.

You totally convinced me. As you all know I'm looking to buy a new car. I currently have a BMW but i'm open to going with another brand. I have been looking at the 2019 BMW X5 which is a bit too costly for me but i'm anxiously awaiting the 2019 BMW 3 series. I also taught about getting a Cadillac CT6 with supercruise but i backed out of it at the last minute.

I can't wait for 2019, especially with the 35k Model 3. Tesla will have FSD and i will be able to go coast to coast while sleeping in my car.
Thanks for opening my eyes and showing me the right way Tesla will achieve general autonomous driving.

I can't wait to buy my Model 3 in Q1 2019 and start sleeping in my car by Q2 2019!

 

[acoste]

Steve Nadig, Daimler's Head Engineer for Mechantronics is not willing to guess that it can replace human driver in 5 or even 10 years.

That despite of its world's first road-licensed self-driving truck in Nevada 3 years ago.

The reporter continued to say the engineering consensus for autonomous trucks is 10, 20, maybe 30 years which will be good for easy long stretch highways but it will still need human for local roads.

What's going on? I thought autonomous technology is either here now or soon but now PBS drops a 30 year time bomb?

There is a difference in mentality of car makers.

Elon Musk released Autopilot in late 2014 and made the people believe this is a self-driving car. This was a reckless step since they knew the car can not handle simple situations, like stopped car in lane, car cuts in front, merging and so on. Camera had very limited recognition capabilities. Results came soon after and taking its toll ever since with numerous major accidents. This in my view is not different from GM's ignition switches where the leaders knew its defect is killing people, but why not keep it this way as long as it is profitable.
Thankfully Elon learned some from that Autopilot experience and decreased the autonomy of the software ever since the first release making sure the driver pays attention.
But didn't learn enough. What he is calling now "Full Self-Driving Capability" is "we believe will be a probability of safety at least twice as good as the average human driver". So this version will still unable to solve a bunch of situation and will still cause accidents.

I would not call anything self-driving that can cause accidents and can't solve certain situations and this is how Steve Nadig thinks as well. That's why the opinions differ a lot.

Google realized that on the way of developing a generic self driving car there is a possibility to release geo-limited versions that at the same time can make some profits. This is still not ready yet, therefore (unlike Elon) they leave the necessary safety checks in place until it is a proven solution.

 

[acoste]

Forward-facing only though. Am struggling to get the point of that.

The point is that lidar is getting closer to being utilized in production cars. First it will be in the large luxurious cars since those can swallow the cost easier and as technology advances the price drops. Lidar needs less computing power. Right now a unit with 250m range and good resolution costs around $8k. Getting closer to reality.

At the same time the camera + radar solution price will increase because the current cameras are not good enough to detect everything, radars give a lot of false alerts and higher resolution cameras plus higher resolution radars will be necessary. Plus the extremely high computing power. They may be able overcome these difficulties in 10 years, but not in the near term.

The reason I think camera (and radar) solution is not the way to go right now is because even if the software is at its top performance (far from it now) it can still be tricked. For example objects in the shadow on the road. Objects on road in rain. Objects on road in the dark. Or if someone wants to be funny, just paints a far from ideal body on the road and all the camera based self driving cars will freak out. A funny shaped pothole can do it as well. Same if someone paints a road on the wall at the end of the street. And same if someone's clothes match the color of the background. There is no distance information either. They may need stereo cameras that will increase the computing power requirements.

 

[Tam]

I applause current Waymo's cautious approach but the following article also raised questions on how fast the technology has progressed as it has abandoned the following 2:

1) no need for human driver
2) Public: is now defined as those who are already in the prior program. 48 hours later, Waymo currently has refused to disclose to the public the numbers of sign-ups and number of rides so far.

Even self-driving leader Waymo is struggling to reach full autonomy

 

[strangecosmos]

Since the early rider program is free, and Waymo One costs as much as Uber/Lyft, why would anyone switch?

 

[Tam]

...why would anyone switch?...

I, myself would love to use an Autonomous Vehicle if it's mine and not from someone else's.

Since current automation is not as aggressive and as efficient as a human is (unprotected left turn, lane changing...) it might be too early to convince customers to choose to pay to a rookie/student driver-like system rather than to a professional human.

Waymo might need to start with paying system with dirt cheap rides first then gradually creep up the price to match with the growing pool of riders.

Maybe Waymo doesn't like that slow ramp up in fees because it wants to make money as soon as possible because of expensive hardware and software and now with paying a human backup driver as well.

It'll be very interesting to see whether there are enough nerds in the region who are willing to pay to ride in a very early automation system.

 

[voyager]

Perhaps it is better to have the vehicle adapt to new technology... instead of expecting that new technology will make the car deal with present day traffic.
Three solid reasons why TNCs (UBER, Waymo) are on the wrong track:

 

[Engr]

It's ironic that after all the hype about lunching *driverless* cars, the *driverless* cars leader still lunched with... a *driver*

 

[mongo]

There is a difference in mentality of car makers.

Elon Musk released Autopilot in late 2014 and made the people believe this is a self-driving car. This was a reckless step since they knew the car can not handle simple situations, like stopped car in lane, car cuts in front, merging and so on. Camera had very limited recognition capabilities. Results came soon after and taking its toll ever since with numerous major accidents. This in my view is not different from GM's ignition switches where the leaders knew its defect is killing people, but why not keep it this way as long as it is profitable.
Thankfully Elon learned some from that Autopilot experience and decreased the autonomy of the software ever since the first release making sure the driver pays attention.
But didn't learn enough. What he is calling now "Full Self-Driving Capability" is "we believe will be a probability of safety at least twice as good as the average human driver". So this version will still unable to solve a bunch of situation and will still cause accidents.

I would not call anything self-driving that can cause accidents and can't solve certain situations and this is how Steve Nadig thinks as well. That's why the opinions differ a lot.

Google realized that on the way of developing a generic self driving car there is a possibility to release geo-limited versions that at the same time can make some profits. This is still not ready yet, therefore (unlike Elon) they leave the necessary safety checks in place until it is a proven solution.

Disagree
AP1 was only AP
FSD didn't exist as a future option until AP2 in 2016 when EAP was also introduced.
Media and ill informed people are the ones erroneously calling Teslas self driving.
Non-AP cars are more deadly than Tesla, just take your hands off the wheel with cruise control on to verify.
Level 4 is self driving in limited situations.
No driver, human or otherwise, is accident proof. Being in an accident does not mean you caused it.

Not every driving situation be solved.

 

[Bladerskb]

Level 4 is self driving in limited situations.
No driver, human or otherwise, is accident proof. Being in an accident does not mean you caused it.

Not every driving situation be solved.

Elon said Teslas would be level 4 in jan 2018, then before may 2018 and now has doubled downed that they will be level 5 in 2019.
Unlike Waymo, Elon actually has a track record of delivering. So i completely believe him. Why else would he double down literally 1 months ago? He knows 2019 is in a-couple weeks. Its not like he's double downed on things months before the due date. An exponential improvement, a step change will happen in acouple of months that will lead to level 5 everywhere in all situations as Elon described.

 

[mongo]

Elon said Teslas would be level 4 in jan 2018, then before may 2018 and now has doubled downed that they will be level 5 in 2019.
Unlike Waymo, Elon actually has a track record of delivering. So i completely believe him. Why else would he double down literally 1 months ago? He knows 2019 is in a-couple weeks. Its not like he's double downed on things months before the due date. An exponential improvement, a step change will happen in acouple of months that will lead to level 5 everywhere in all situations as Elon described.

Elon has projected FSD, but has not said the cars on the road that people buy are currently FSD, that's what I'm saying.

P.S. You ok? Either your account was hacked, you recently had a stroke, started working for Tesla, walked to Damascus, or have been dropping a "/s" in your posts.