Waymo (Google's Self Driving Car Company)

[Cashmere]

Starting today, the drive for autonomy is called Waymo, a standalone company under the Alphabet corporate umbrella. And that means it’s time to take the technology to market.

“We’re a self-driving car company with a mission to make it safe and easy for people and things to get around,” says Waymo CEO John Krafcik. What that means, exactly, is still an open question: Krafcik mentioned ridesharing, trucking, logistics, even selling personal use vehicles to individual consumers.

• aggressive plans to bring fully self-driving cars to market, with launch dates ranging from next year to 2021. Meanwhile, pressure to prove Google’s X projects could deliver on their promises mounted within the company, and a series of executives abandoned Google’s project, including longtime technical lead Chris Urmson, who was reportedly unhappy with Krafcik’s leadership.
  
  As Uber and nuTonomy welcomed the public into self-driving cars in Pittsburgh and Singapore, respectively, Google’s car team ignored questions about how and when it would commercialize its tech. Reports claimed it was scaling back its ambitions. The effort started to look like an also-ran. Launching Waymo, at least in the public eye, puts the robot-car division back in the race.
  • Boston Consulting Group pegs at $42 billion a year by 2025. So putting Steve Mahan in the car feels like a publicity stunt—look at our good deeds!—until you talk to the man.
    
    Mahan started losing his vision at the age of 37, in 1990, due to a rare genetic disease called nanophthalmos. He was blind in one eye within a few years, and lost sight in the other after a laser treatment in 2004. (Doctors salvaged about 10 percent of his vision, but expect him to lose even that, eventually.) He stopped driving soon after, and for the past decade has relied on generous family and friends, and onerous paratransit services, to get from his suburban home in Morgan Hill, California, to anywhere else. Losing that independence, he says, was traumatic.
    
    Hitting the green “go” button in the Google pod washed away some of that anguish, at least for a moment. “I was given the opportunity to be the man that I was before, for the time that I spent in that car,” Mahan says. “It just let me be a whole person again.”
    
    Mahan had ridden in Google’s car twice before. In 2012, he rode with an engineer and police escort; in 2013, in a closed parking lot. Last year’s Austin expedition was his first solo trip, and the closest approximation to what an autonomous car service could offer. “This trip was about showing that we can actually take that capability and turn it into something that’s ready to go out into the real world,” says Nathaniel Fairfield, who runs Waymo’s planning, control, and routing team, and has been on board since the project launched.
  

https://www.wired.com/2016/12/google-self-driving-car-waymo/

 

[Cashmere]

Waymo’s 100 autonomous Chrysler minivans are here



Chrysler has completed the 100 autonomous Pacifica minivans that will join the Waymo (née Google) fleet in early 2017. The vans, which are plug-in hybrid variants with Waymo’s self-driving hardware and software built in, are part of a partnership between Fiat Chrysler (FCA) and Waymo that was announced earlier this year.

Waymo CEO John Krafcik said last week that his company is not interested in “making better cars.” Instead, it wants make “better drivers.” That’s why Waymo has partnered with FCA, which has not been as active in showing off its autonomous chops as other major automakers like Ford and GM.

“With this great new minivan on the road in our test markets, we’ll learn how people of all ages, shapes, and group sizes experience our fully self-driving technology,” said Krafcik in a statement. Waymo had been testing a number of prototype Pacificas, but these vehicles appear to have much tighter integration of Waymo’s self-driving hardware than those prototypes.

The partnership between Waymo and Chrysler saw engineers from both companies work closely to “rapidly and robustly integrate the vehicle electrical and control systems with the fully self-driving system,” said a Fiat Chrysler spokesperson Berj Alexanian. Most of Waymo’s 2 million miles of autonomous testing have been in a fleet of Lexus RX SUVs with Google’s self-driving hardware added on.

FCA and Waymo engineers modified the Pacifica’s electrical, powertrain, chassis, and structural systems to ready the van for Waymo’s self-driving tech. The two companies have been testing prototype Pacificas at FCA’s Chelsea Proving Grounds in Michigan and Arizona Proving Grounds in Yucca, Arizona, as well as at Waymo facilities in California. The trials included 200 hours of “extreme-weather testing.”

“Waymo learned first-hand what goes into an automaker’s vehicle development process, such as optimizing weight distribution to ensure a comfortable driving experience and durability testing in extreme-weather conditions,” said Alexanian.

Waymo and FCA may even be working together on an autonomous ride-sharing service that could launch as soon as late 2017, according to a Bloomberg report, likely with these autonomous vans at its core.

The autonomous Pacificas should hit the streets early next year, likely in existing Waymo testing markets in California, Washington, Arizona, and Texas.

http://www.theverge.com/2016/12/19/14003642/waymo-google-autonomous-chrysler-pacifica-minivan-photos

 

[chonkoa]

Yes the minivans are here just like the glasses from google.

 

[Cashmere]

Waymo is first to put fully self-driving cars on US roads without a safety driver

Going Level 4 in Arizona

Waymo, the autonomous vehicle division of Alphabet, Google’s parent company, reached an important milestone recently: since mid-October, the company has been operating its autonomous minivans on public roads in Arizona without a safety driver — or any human at all — behind the wheel. And starting very soon, the company plans to invite regular people for rides in these fully self-driving vehicles.

The news that Waymo’s vehicles have been on public roads with no human in the driver’s seat was announced today by the company’s CEO John Krafcik at a tech conference in Lisbon. The announcement comes on the heels of Waymo’s decision to invite a group of reporters to visit Castle, a 91-acre facility in California’s Central Valley that it has been using as a training course for its self-driving vehicles. At the time, Krafcik declined to provide an exact timetable as to when it would begin testing fully self-driving vehicles on public roads. Little did we know at the time, they were already doing it.

Early Rider program, which has been in operation in Chandler since last April, will be the first to experience the new technology.

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

But it most certainly signals a growing level of confidence by Waymo, and, by extension, Alphabet, in the technology developed by its engineers over the last eight years. It represents a raising of the stakes for the rest of the industry that sees full, Level 4 autonomy, in which a vehicle is capable of driving itself, with no human behind the wheel, in most environments and road conditions, as the ultimate goal.

In the year since it spun off from Google, Waymo has moved aggressively to test its cars on public roads. The company has certainly been testing self-driving cars on public roads longer than the broader tech and auto industries, and it has access to Google’s machine engineering talent, making it a formidable competitor to companies with decades of car-building experience.

But Waymo isn’t going it alone. The company has partnerships with Fiat-Chrysler, Lyft, and Avis, even while it’s manufacturing its own sensors and hardware to reduce dependence on suppliers. In his speech, Krafcik enumerated the many advantages that Waymo’s vehicles have over its competitors.

“This is the most advanced vehicle we’ve developed to date,” he said. “Everything in it is designed and built for full autonomy. Our combination of powerful sensors gives our vehicles a 360 degree view of the world. The lasers can see objects in three dimensions, up to 300 meters away. We also have short range lasers that stay focused close-up to the side of the vehicle. Our radars can see underneath and around vehicles, tracking moving objects usually hidden from the human eye.”

Its last report to California showed that Google’s disengagements dropped precipitously.) Waymo recently released its first safety report, a glossy, 41-page document that details its testing regime that involved 3.5 million miles driven on public roads and 2.5 billion miles driven in simulation. The report, which was released voluntarily, earned the company an atta boy from US Secretary of Transportation Elaine Chao.

Chao’s department is scrambling to amend its voluntary autonomous vehicle guidelines to include buses and trucks, the latter of which Waymo is also working on. And Congress is weighing legislation that would give companies wide latitude to deploy, test, and make money off of self-driving vehicles. But most companies are still moving cautiously, cognizant of the fact that most people are skeptical or flat-out distrust autonomous technology. They say they plan to keep their safety drivers behind the wheel until 2020 at the earliest.

But not Waymo.

Link: Waymo is first to put fully self-driving cars on US roads without a safety driver

 

[Giannis]

Just for fun, I've recently started noting how many near crash experiences I have per day due to other drivers' inability to face basic rules. I average about 4 per day over the last month, with a maximum of 12.

In this world I wouldn't trust a self driving car.

 

[Sunny]

Just for fun, I've recently started noting how many near crash experiences I have per day due to other drivers' inability to face basic rules. I average about 4 per day over the last month, with a maximum of 12.

In this world I wouldn't trust a self driving car.

Wouldn't that be more a reason to trust an self driving car?

 

[Giannis]

Wouldn't that be more a reason to trust an self driving car?

No, I don't think so.

Surviving the roads of southern Greece requires a set of extremely good reflexes, patience, luck and guessing skills. A self driving car would be good only in the first part.

 

[Centurion]

No, I don't think so.

Surviving the roads of southern Greece requires a set of extremely good reflexes, patience, luck and guessing skills. A self driving car would be good only in the first part.

Isn't the role of AI to learn, adapt and predict behavior? If you can then an algorithm should be able to do it alot better. The radars and cameras are also designed to mathematically record patterns of behavior of other cars, and even make note of patterns specific to days if the week of hours of the day.

Google stands a good chance of championing autonomous driving. Hardware is the easy bit, software is the hardest aspect and few companies do software, data science or algorithms better than Google.

 

[EnI]

Self-driving cars are not really self-driving. They will be @ Level 5. But that stage is at least a decade away if not more.

But first we need 5G network for that, smart traffic signs lights, more accurate GPS positioning, LIDARs with dedicated processing units, in-cloud AI processing AD and constantly updating the car's system, car-to-car communication etc etc. So, it's not all up to carmakers & IT companies ... it has to do a lot with telecommunication companies (telecoms) as well; states & local communities (smart cities; smart traffic signs & lights); GPS providers; LIDAR makers etc.

All lower levels & stages are purely experimental, beta ... useless. Just enhanced driving safety assistance systems. Not true autonomous driving systems.

Level 4 is autonomous only in (semi)controlled, known environment. Mostly multi-lane highways.

The systems for urban, inner-city usage are much more complicated - dealing with (moto)cyclists, pedestrians, still traffic etc etc - needing much more powerful hardware & software to process all the date in real time. Not here though in series production.

Also - to not to look like an ambulance vehicle with a rotating light on the roof - cars will need at least 3 (most probably 5 or more) LIDARs located in various parts of car's body (bumpers, grille, fenders, columns etc). One central LIDAR on the roof (as seen on Waymo prototypes) is not acceptable for car companies - it hurts the looks, aerodynamics, roof practicality, it increases vehicle's center of gravity etc etc. Perhaps it's ok for hailing / taxi companies & commercial vehicles but not for privately owned passenger vehicles. And accommodating 5 LIDARs will be VERY expensive until the tech goes mainstream (like eg ultra-sound parking sensors etc).

Tesla's AD "solution" without LIDARs (and more stereo-cameras instead) has been proven useless & inadequate by ALL the AD researchers bar Tesla. Mind Mobileye and Tesla end the collaboration due to that particular issue. And it seems Tencent is on the same way: now developing its own LIDA-based AD system, despite cooperating with Tesla on Tesla's stereo-camera based system. Elon insists LIDARs are not necessary (and way to complicated & expensive solution - hardware & software wise), and sticks with stereo-camera based solution. Mind Tesla is the only company that does that!!!! Even deceiving it's customers the optional AutoPilot hardware in Tesla 3 cars is AD Level 5 ready - when proper software is due.

So, currently AD is still very much in the experimental, developing phase. And it will stay there for at least a decade.

Not to mention legislative hurdles ... The legislation is still not ready for Level 5 AD. So until it is AD is more or less illegal (bar for experimental & developing purposes - which require special permissions & licences by authorities).

 

[Sunny]

No, I don't think so.
Surviving the roads of southern Greece requires a set of extremely good reflexes, patience, luck and guessing skills. A self driving car would be good only in the first part.

... so that leaves patience, luck and guessing skills.

Patience - I am sure you agree machines don't suffer from impatience, by corollary they surely have more patience than humans.
Luck - Being a man of science, I am also sure you agree that has nothing to do with who/what is at the wheel.

So that leaves guessing skills - or rather predictive ability. Willing to wager machines will soon be comprehensively much better at this too.

Isn't the role of AI to learn, adapt and predict behavior?

That is is the key part. I think there is a gross misunderstanding here of how "AI" works. People think it is a massive computer program with a massive "if" block, which lists every possible event and outcome. And by any chance, in the real world, if the program encounters an event that it is not programmed to handle, it will fail. But that is not how it is.

In a traditional computer program, yes, you give it a set of rules. The program then applies the rules to a given input to produce an output. The output is obviously, only as good as the rules you program into the computer. If you don't program a rule to handle when a traffic light flashes yellow, the output won't be good. In an AI (or ML)program, it is other way around, you give it the desired output and let the computer learn (or train) what rules to apply to get that desired output.

I would highly recommend reading this to just fathom the speed of what is happening in the field of AI research - AI versus AI: Self-Taught AlphaGo Zero Vanquishes Its Predecessor

The gist - last year, Google's Deepmind AlphaGo (a program that used machine learning to train from human games) beats the Go world champion 4-1. This is the first time in 2500 years a computer program has beaten a top rated human player in the game of Go with out an handicap. This year, AlphaGoZero (the next version) that trains by playing against itself, after just 3 days of training itself, beats AlphaGo 100-0.

The AI That Has Nothing to Learn From Humans

 

[EnI]

The problem with AI is that it is good to handle situations when rules are known & set. AND ... most importantly(!!!) : every participant fully complies to those rules!!! Eg. the games etc.

Real life is completely messed up environment. There are tons of different, specific, special situations going on constantly. Perhaps too specific & too little of them happening for AI to make a pattern out of it. And even when it does it is limited by basic rules - eg. what to do in dangerous situations - and can't eg perform theoretically the most optimal move ... meaning: it can't create it's own (more optimal) rules. Just like AI player @ game playing. It's set by basic rules.

But as said: in real life there are "players" (drivers, pedestrians, cyclists, motorists, animals, objects, weather events etc) who don't always follow the basic rules but often break them. Eg. a pedestrian waiting to cross the street, waiting for a green light ... but then he / she decides to cross it at red light anyway. What should AI Autopilot do? Assume that at any time, any situation, any pedestrian, any animal, any driver, any cyclist can break the rules??? No. There has to be another filter layer. But how to create it? What to monitor? Whit which sensors?

Human brain - paired with incredible visual & audio sensors: eyes & ears - is a genial apparatus. It can monitor & percieve and so predict the behaviour of traffic participants. Eg. checking & observing what those participants are doing - are they fully alert & concentrated on what's going on in the traffic; are they distracted somehow; have they detected the situation / the danger. Etc etc.

Our brain together with eyes can eg. check pedestrian head & eyes position & evaluate his/her activity (eg. looking at phone; listening to the music due to earbuds in the ears etc) to predict the his/her behavior and anticipate the next possible move. AI can do that (yet) due to lack of proper precise (high-resolution) sensors.

Traffic AI has to be supported with various & many very precise sensors, hardware to be able to evaluate all the situation properly. Eg. especially in the crowded urban inner city streets. Where the situation is usually erratic & chaotic. Eg. on highways the situation is much more controlled - there's why everybody are more concentrated on developing AI for highway driving mostly. Because it's simpler - software & hardware wise.

To make a great autopilot for crowded urban , inner city roads a car has to be equipped with tons of various high-res sensors, AND a proper AI software & hardware (eg. in the cloud). And right now such solution & feature is ridiculously expensive. And it still lacks some parts: eg. faster mobile internet transfer (it will be available with 5G) and more precise positioning (more precise GPS systems also coming in the near future) and VERY detailing, constantly updating mapping (also under development).

AI will work best when there will be more AI driven & controlled traffic participants on the roads. Not only motor vehicles but also pedestrians & cyclists as well: connected to traffic info system eg. via earbuds or some sort of goggles ... making people in traffic more controlled & predictable. Eg. phone front cameras (when looking at them) & goggle eye sensors will monitor our eye & head movement so the AI will predict our behaviour.
Also the monitors & cameras in trafic lights & traffic signs etc will monitor the traffic (especially the pedestrians, cyclists, animals, other moving objects etc) and complement the in-car sensors and the sensors in goggles, phones, earbuds etc to feed info to the central traffic AI algorithm which will asses the situation in real time (imagine that computing power and that constant data transfer!!!!!!!!!) and come out with proper command or action.

It's a very complex & complicated solution. It can work though. It will work though. But there is still a huge lack of proper hardware - and consequently software. But the theoretical solutions are there & ready. Just waiting for proper hardware to be put in series production & ready for mainstream use.

 

[Sunny]

Eni, no offense, but I don't think you grasp how AI or rather specifically convolutional neural networks work. Neural network and deep learning and is being used in many fields where there are no rules, in fact, fields where the agents deliberately break rules and even cover their paths. If you have a CS/math background I would advise taking some AI/ML courses. Right now, you have a simplistic, preconceived notion of what AI is and thinking that is where the horizon ends - classic Dunning Kruger effect.

 

[Centurion]

Traffic AI has to be supported with various & many very precise sensors, hardware to be able to evaluate all the situation properly. Eg. especially in the crowded urban inner city streets. Where the situation is usually erratic & chaotic. Eg. on highways the situation is much more controlled - there's why everybody are more concentrated on developing AI for highway driving mostly. Because it's simpler - software & hardware wise.

Mind you that there are Fintry companies using AI software for day and short term trading. This is an example of AI being used to replace a human's work in a chaotic and emotionally(sentiment, hype, fear) laden environment.

Although you are raising a good point about automotive AI initially focusing on environments of more predictable behaviour e.g motorways, there are big advances being made in deep learning and neural networks as @Monster said.

The ultimate conquest for self driving cars is to operate autonomously in environments where rules are constantly broken and this will require terabytes of data stored locally. I don't think the technology will be perfected in the next 3 year but more in 5 years.

 

[Giannis]

Isn't the role of AI to learn, adapt and predict behavior?

We've yet to see an AI algorithm that can predict that the car in front, with no turn signal used, will indeed turn right.

Luck - Being a man of science, I am also sure you agree that has nothing to do with who/what is at the wheel.

There's a machine workshop out of the city that I drive to quite often. You need to take a narrow mountain road to get there. The road has 4 blind turns, following the curvature of some small hills. There's no way to see if anyone is coming from the other side. My approach is to drive extremely slow (5 km/h) and use the horn liberally.

Once, a blithering idiot with a modded Citroen Saxo VTS came out of one such corner doing about 80 km/h and luckily I was just 50m away from the bend, enough distance to come to a halt and almost touch the end of the tarmac (followed by a 20m drop down the mountain - no safety barrier). If I was in the bend, what would the AI do?

Eni, no offense, but I don't think you grasp how AI or rather specifically convolutional neural networks work. Neural network and deep learning and is being used in many fields where there are no rules, in fact, fields where the agents deliberately break rules and even cover their paths. If you have a CS/math background I would advise taking some AI/ML courses. Right now, you have a simplistic, preconceived notion of what AI is and thinking that is where the horizon ends - classic Dunning Kruger effect.

These days I'm reading into neural networks for use as a mean of predicting failure of mechanical components. I think I have a vague understanding of what self teaching the algorithm means.

Though, I live in the poor south of Greece. Driving here is nothing like the civilized environment of western Europe. Most of the time, we even struggle with the definition of road, let alone driving rules.

 

[Centurion]

Most of the time, we even struggle with the definition of road, let alone driving rules.

Hence why the ultimate test for AI will be self driving in Mumbai, rural Russia and Mogadishu . To do so successfully AI will occasionally have to be aggressive and blocking other drivers and sticking its nose through gaps, or else it will be bullied and not get anywhere.

 

[Sunny]

We've yet to see an AI algorithm that can predict that the car in front, with no turn signal used, will indeed turn right.

What if self driving car read the plates and from previous patterns (shared among other sd cars) knew the probability the car would make the next right? What if AI knew that at 8AM at that particular intersection the probability of a mini van making right turn was x% cause it is school time and there is a school down the right turn? What if it identified the vehicle as a bus and read the bus number knew the bus route involved making a right turn? What if it read the name of business name at the back of the truck and knew the business address and knew the probability the truck would make right? All just me pulling stuff from my back side - but you get the picture, the possibilities are endless. And the best part, all this would be shared knowledge among the self driving cars that is constantly and instantaneously updated.

If I was in the bend, what would the AI do?

What would you do? I don't think self driving cars is panacea to all vehicular accidents. There will still be accidents. So I don't know where this expectation is arising that unless self driving cars make every accident a thing of the past, it is some how not good enough. That would a tragic human folly. This article I linked to previously actually talks about this - A robot is coming for your job

...Ultimately, our belief systems are simply out of sync with the data. Self-driving cars, which have performed flawlessly during their millions of hours logged on real streets, still have a driver at the wheel as Uber launches its first fleet. Statistically speaking, passengers are in much more danger with a human operator involved in any capacity. After all, human errors literally kill over a million people a year, whereas the statistical likelihood of dying from a self-driving car is like falling off a building and being struck by lightning on the way down. Nonetheless, an operator sits in the driver’s seat of each of Uber’s self-driving cars, hovering their hands over the wheel.

This is ultimately a metaphor for the whole problem that is restricting our progress: We would rather let millions die in traditional ways than risk even one person dying in a way that is new and scary. So I propose that if you find yourself uncomfortable with any job that will soon be automated by AI, you should step back and try and remember an experience with a human underperforming in the same capacity, and ask yourself: “Could a robot really do much worse?”

These days I'm reading into neural networks for use as a mean of predicting failure of mechanical components. I think I have a vague understanding of what self teaching the algorithm means.

One of the best resources is the Deep Learning course by AndrewNg on Coursera - Neural Networks and Deep Learning | Coursera. Highly recommend it and it is free!

 

[Wolfgang]

Mike Jackson used to lead MBUSA in the late 1990s or so... Now he likes Waymo?


AutoNation, Waymo and a 6-year courtship

In the fast-moving world of technology disruption, a six-year gestation seems extreme. But in one case, that's what it took.

On Nov. 2, AutoNation said it was forming a partnership with Waymo to service Waymo's fleet of self-driving Chrysler Pacifica Hybrids. But the origin of the deal between the nation's largest new-vehicle retailer and Google's self-driving car affiliate stretches over the last six years.

Every 12 to 18 months during that time, AutoNation CEO Mike Jackson has visited Google X, Google's r&d facility in Mountain View, Calif.

"I really admired the discipline and approach they were taking, especially the epiphany of realizing [that] asking a driver to supervise the autonomous system is not really a correct expectation," Jackson told Automotive News. "Autonomous systems really need to be capable of doing the job without human supervision."

In late 2016, Waymo was officially established as a sister company of Google with John Krafcik, a friend of Jackson, as CEO. Krafcik and Jackson discussed the sophistication of Waymo's vehicles. The Pacificas in Waymo's fleet are outfitted with Level 4 self-driving technology, which means they drive without human intervention in geographically designated areas.

Extending the life cycle of those Pacificas "was essential for the ultimate business," Jackson said. "That level of investment in technology in the vehicle would require that the vehicle stay in service for hundreds of thousands of miles for the economics to make sense."

Jackson: Admires Waymo approach

Jackson recalled telling Krafcik, "We [have cared] for 40 million vehicles over the years here at AutoNation. We have tremendous expertise in proactive life-cycle management." And that, because of "our scale, technical expertise and that we're in agreement that your approach is the correct one, we should partner."

In addition to the Jackson-Krafcik meetings, Waymo confirmed that it hosted the entire AutoNation’s board in January to explain the tech company’s vision and strategy.

Under the deal, AutoNation initially will work with Waymo's autonomous vehicle program in the Phoenix and Northern California markets. The agreement to provide vehicle maintenance and repairs to the fleet is not exclusive; another large dealership group theoretically could work with Waymo, too. AutoNation will use its franchise stores, AutoNation USA used-only stores and other centers to care for the vehicles.

Waymo's mission to develop a fully autonomous vehicle was a significant draw for AutoNation, Jackson said during AutoNation's third-quarter earnings call this month.

"They had the whole approach to autonomy where you have to under promise and over deliver rather than over promise and under deliver when it comes to autonomy and safety," he said. "It's almost against human nature that you have to pay strict attention and be prepared to intervene. So, they said they don't need 99.9 percent perfection; they need 100 percent perfection."

Without a fully autonomous system, "You're saying, 'Enjoy and trust this system, [but] never take your eyes off it,'" he said. But with Waymo, "If the consumer truly trusts it, then they stop paying attention."

If consumers don't trust the technology and pay strict attention throughout each ride, they will think, "'I might as well drive the car myself.' And I think this twilight zone is what the bigger industry is going to struggle with," Jackson said.

The partnership will better position AutoNation for the future of the industry, which Jackson says will involve car-sharing, electrification and autonomous vehicles. "We think we can play a vital role in that," he said. "And in my view, Waymo is the company I want to do that with."

Source: Automotive News

 

[Cashmere]

Tesla's AD "solution" without LIDARs (and more stereo-cameras instead) has been proven useless & inadequate by ALL the AD researchers bar Tesla.

Not sure where you got this from but it is incorrect. Several universities including MIT teamed up together to research if self driving is possible using only cameras and their research concluded that LIDAR is not a necessity.

Source: https://arxiv.org/pdf/1708.09839.pdf

Just because Tesla is the only car manufacturer that is not pursuing LiDAR, does not mean they're wrong.

 

[Cashmere]

Waymo is getting ready for mass rollout of Level 4 autonomy:

"With the world’s first fleet of fully self-driving vehicles on the road, we’ve moved from research and development, to operations and deployment," said John Krafcik, CEO of Waymo. "The Pacifica Hybrid minivans offer a versatile interior and a comfortable ride experience, and these additional vehicles will help us scale."

Waymo strikes a deal to buy ‘thousands’ more self-driving minivans from Fiat Chrysler

Link: Waymo strikes a deal to buy ‘thousands’ more self-driving minivans from Fiat Chrysler

Interesting comparison of their buying pattern for the last couple years:

2016: Waymo orders 100 vehicles
2017: Waymo orders 500 vehicles
2018: Waymo orders 2,000 vehicles (lower estimate based on article above)

 

[Centurion]

^^
Additional cars are needed for more testing and data collection.