Levandowski flips the table on L4 narrative

[jimmy_d]

Anthony Levandowski - arguably the most controversial and one of the most experienced program leaders in the AV business - has reemerged to start a new venture in the AV space and has (below) dropped a commentary on medium refuting the dominant narrative in the AV space.

Levandowski has changed his public stance from supporting the 'straight to L4 / transportation as a service' approach to 'ADAS first, start with L2'. As part of this he seems to be supporting learning methods (mainly NNs) over human written methods, and he's moving away from advanced sensors (read here: LIDAR) as being useful to development. He's even using Elon's "crutch" terminology.

So while he is not calling out Tesla by name he seems to be declaring full throated support for Tesla's approach. This is a pretty dramatic reversal for someone who was instrumental in earlier programs to lead with L4/transportation as a service (notably, Waymo).

His company plans to produce a 'copilot' for trucks available as add-on hardware.

I expect there will be pushback on this revision of the narrative. The conventional wisdom that Waymo/Cruise/Uber type approaches are the 'leader' might start to unravel.

Pronto Means Ready – Pronto AI – Medium

 

[electronblue]

@jimmy_d He has a new company and a new thing to sell. No wonder he changes his tune. The software and cheap sensor approach makes sense for a startup behind leaders too.

That said it seems confirmation that indeed the approach here — and likely for Tesla too — is Level 2 driver’s aid first. Keep the driver responsible not the car. The driver is their crutch.

Making a car responsible autonomous system is a very different task.

 

[Mo City]

To be the best means overcoming the biggest challenges that self-driving technology faces head on. Our approach? Much better software. After all, the best and safest drivers don’t necessarily have the best eyes. They have the best brains and the most experience. We are building neural networks from the ground up that combine experience-based AI, end-to-end deep learning, and crowdsourced data with advanced computer vision to deliver a highly scalable and flexible driving stack. Nobody else is doing this.

Yeah right.

 

[electronblue]

To be the best means overcoming the biggest challenges that self-driving technology faces head on. Our approach? Much better software. After all, the best and safest drivers don’t necessarily have the best eyes. They have the best brains and the most experience. We are building neural networks from the ground up that combine experience-based AI, end-to-end deep learning, and crowdsourced data with advanced computer vision to deliver a highly scalable and flexible driving stack. Nobody else is doing this.

Yeah right.

Well, not quite. Credit where credit is due: If they are really doing ”end to end” and mean what that usually means it would certainly be something nobody else is known to be doing beyond tech demos.

Then again it could be just brave startup talk to get financing and a lucrative exit at some point.

 

[jimmy_d]

To be the best means overcoming the biggest challenges that self-driving technology faces head on. Our approach? Much better software. After all, the best and safest drivers don’t necessarily have the best eyes. They have the best brains and the most experience. We are building neural networks from the ground up that combine experience-based AI, end-to-end deep learning, and crowdsourced data with advanced computer vision to deliver a highly scalable and flexible driving stack. Nobody else is doing this.

Yeah right.

I am assuming he excludes Tesla's effort there because he believes (or maybe he knows?) that they aren't doing end-to-end as part of their effort right now. Or were you thinking of someone besides Tesla? It seems pretty likely that *someone* else is doing it. For instance, Comma AI might be doing all of those.

 

[DanCar]

No body is a leader in my opinion. Waymo certainly isn't. They are too afraid of killing or injuring someone, so they won't take any risks. Driverless tech won't work with zero risk. In my opinion Tesla is the leader because:
1. Tesla is willing to take risks.
2. A simple system often wins over a complicated system. No lidar makes the system simpler.

What Levandowski says makes sense to me. I like his approach but there is still the obstacle of overcoming unexpected situations. Which means having a human ready to take over, perhaps via remote control, will still be needed for a long time.

 

[CarlK]

This is very interesting. He's not the only one beside Tesla that is endorsing this approach though. George Hotz of Comma.AI, another super smart software guy, has been doing exactly the same thing, camera and NN, since at least 2015'. Here it raises another interesting question. Would Levandowski go this route now if he has stayed with Waymo? I would say no. It's really hard to re-direct a large organization even for a pretty fast moving company like Google. That's why new disruptive technologies are usually dominated by new comers. In the different arena this is what's happening in electric vehicles too.

 

[strangecosmos]

Key quotes from Levandowski’s Medium post.

On Level 4 and 5 autonomy:

“...the reason why nobody has achieved this level of functionality is because today’s software is not good enough to predict the future. It’s still nowhere close to matching the instincts of human drivers, which is the single most important factor in road safety.”​

On lidar:

“...traditional self-driving stacks attempt to compensate for their software’s predictive shortcomings through increasingly complex hardware. Lidar and HD maps provide amazing sensing and localization of the present moment but this precision comes at great cost (with respect to safety, scalability and robustness) while yielding limited gains in predictive ability.”​

Overall assessment of the industry:

“Put simply, the self-driving industry has gotten two key things wrong: it’s been focused on achieving the dream of fully-autonomous driving straight from manual vehicle operation, and it has chased this false dream with crutch technologies.”​

On Pronto AI’s approach:

“Over the past three years, amazing gains in machine learning and a new breed of tensor processing hardware have made it possible to pursue a different, ultimately much more promising, path toward solving the self-driving challenge. That’s what my new company — Pronto — is all about.​

...Our approach? Much better software. After all, the best and safest drivers don’t necessarily have the best eyes. They have the best brains and the most experience. We are building neural networks from the ground up that combine experience-based AI, end-to-end deep learning, and crowdsourced data with advanced computer vision to deliver a highly scalable and flexible driving stack. Nobody else is doing this.​

We are not building technology that tells vehicles how to drive. Instead, our team of engineers is building tech that can learn how to drive the way people do.​

Our tech does not shy away from the rich complexities of real-world driving. Through better prediction and decision-making software, we are able to navigate previously vexing “edge cases,” such as very low light, direct sunlight glare, heavy rain, snow, construction zones, etc. in a safe, scalable, and repeatable manner on a wide variety of highways without mapping them.”​

On Pronto AI’s first product, Copilot:

“The first step is to deliver a commercially-viable ADAS product that makes driving safer for everyone. It augments the driving experience by reducing the cognitive workload for drivers, allowing them to focus their full attention on monitoring the road ahead. The market that we believe makes the most sense to engage first is the commercial trucking industry...”
​

Demo video:

 

[jimmy_d]

This is very interesting. He's not the only one beside Tesla that is endorsing this approach though. George Hotz of Comma.AI, another super smart software guy, has been doing exactly the same thing, camera and NN, since at least 2015'. Here it raises another interesting question. Would Levandowski go this route now if he has stayed with Waymo? I would say no. It's really hard to re-direct a large organization even for a pretty fast moving company like Google. That's why new disruptive technologies are usually dominated by new comers. In the different arena this is what's happening in electric vehicles too.

Regarding the man himself - I was refreshing my memory of his history and found this recent article which I found quite illuminating. https://www.newyorker.com/magazine/2018/10/22/did-uber-steal-googles-intellectual-property

It includes a lot of interesting background on Levandowski and his time at Google. I'd say it thoroughly supports your view.

 

[jimmy_d]

Key quotes from Levandowski’s Medium post.

Demo video:

The choice of sound track is rather interesting.

 

[electronblue]

The cross-country drive is an impressive achievement for a young startup. Of course Delphi already did the same years ago, but still it is more than Tesla has been able to show. I certainly appreciate that.

What is most illuminating here, though, is the literal reversal on car responsible driving angle. While Elon Musk started by talking about Level 5 capable hardware, both Comma.AI and now Pronto.AI — startups from the same autonomous school of thought as Elon Musk with equally vocal leaders — are basically admitting their approach will push back the advent of car responsible driving for a long time in exchange for seeking a vision-only solution. They are not making autonomous cars, they are making driver’s aids at this stage.

They are using the driver as the crutch which is disappointing for anyone hoping to read a book in the driver’s seat, on the highway, anytime soon.

Finally one thing that needs to be remembered is: Levandowski is marketing his new startup, very likely in the hopes of a lucrative exit — after for years being employed by others and seeing the fruits of that labour go into their pockets. A big part of any bravado — from any company — is marketing. This is no different. Levandowski has a thing to sell and he is selling it. Good for him, but we need to consider this when assessing the arguments.

 

[electronblue]

This is very interesting. He's not the only one beside Tesla that is endorsing this approach though. George Hotz of Comma.AI, another super smart software guy, has been doing exactly the same thing, camera and NN, since at least 2015'. Here it raises another interesting question. Would Levandowski go this route now if he has stayed with Waymo? I would say no. It's really hard to re-direct a large organization even for a pretty fast moving company like Google. That's why new disruptive technologies are usually dominated by new comers. In the different arena this is what's happening in electric vehicles too.

The goal is very different here though. Waymo, MobilEye, Audi, BWM etc are working on car responsible driving. Levandowski, Comma.AI and actually many fear also Tesla is basically working on driver’s aid — all the talk of car responsible driving from the latter group has basically subsided in recent times. They don’t want to talk SAE Levels because they do know they will need the driver as the crutch for their approach likely for a long time.

When will these two different approaches intersect and to whose benefit? Good question.

 

[strangecosmos]

Regarding the man himself - I was refreshing my memory of his history and found this recent article which I found quite illuminating. https://www.newyorker.com/magazine/2018/10/22/did-uber-steal-googles-intellectual-property

This article makes it sound like it sound like Levandowski was never shown to have taken any trade secrets from Waymo, and that Waymo decided to settle the case because they were afraid they would lose. The judge said that some of the things Waymo were claiming were trade secrets were actually general principles of science and engineering.

It also turns out the lidar patent that Waymo was suing Uber for infringing was invalid all along: Vigilante engineer stops Waymo from patenting key lidar technology

The article gives me the impression that trade secret laws and litigation are kind of BS in general; they are overly broad and subjective, and they have a chilling effect on free market competition among companies and workers. At what point does the definition of “trade secrets” expand to include all the skill, experience, and knowledge that engineers have developed after spending time working on a problem? When does owning “trade secrets” effectively just become owning the engineers themselves?

The article makes it seem like a big part of Waymo’s motivation for suing Levandowski — maybe the primary motivation — was to reduce Uber’s competitiveness with Waymo, and to make an example out of Levandowski, to scare off other Waymo or Google employees from leaving. It sounds like Google/Waymo wanted to behave anti-competitively first, and then instructed its lawyers to go digging for anything they could find to gin up a legal case — which eventually turned out to be mostly, or entirely, without merit. It doesn’t sound like they were sincerely trying to protect their intellectual property, or that their main motivation was any actual perceived merits of the case.

I don’t know if Levandowski is a person I would trust. The article says he engaged in workplace bullying, and, at least as the article tells it, remained defensive rather than showing remorse or concern after getting his boss/coworker into a crash that caused him to “injure his spine so severely that he eventually required multiple surgeries.” But bullying and defensiveness aren’t outlier, villainous behaviours. They are pretty common among all sorts of people from all walks of life. A lot of Silicon Valley workplaces sound unprofessional, abusive, and dysfunctional — like a lot of non-tech workplaces, and relationships, and families. Is Levandowski really so different in that regard?

In particular, I think boys and men are typically taught that most emotions other than anger are weak and feminine, and that anger is masculine (conversely our culture has a real problem listening openly to women’s anger, including when it’s legitimate). Boys and men are also taught that the way for them to get love is to be socially dominant, to be “alpha”, to be cool (not warm) and tough and strong and impressive, to seek status and success and money and power. So, is it a surprise that there are a lot of men expressing anger and seeking dominance?

 

[Engr]

They are using the driver as the crutch which is disappointing for anyone hoping to read a book in the driver’s seat, on the highway, anytime soon.

Waymo uses safety "drivers" also. Is that still a crutch or a strength?

Nobody is pushing L4 systems away. After ten years of development even Krafchick admits true L4 will still take a long time to achieve. So what's the solution?
Waymo, Autopilot, Pronto etc. They all require an attentive driver, only difference is in case of Waymo, the drivers have "safety" title.
Waymo is working on phasing out the safety drivers. But, what's the difference between that and Tesla's work on making your car achieve FSD?

 

[electronblue]

Waymo uses safety "drivers" also. Is that still a crutch or a strength?

Nobody is pushing L4 systems away. After ten years of development even Krafchick admits true L4 will still take a long time to achieve. So what's the solution?
Waymo, Autopilot, Pronto etc. They all require an attentive driver, only difference is in case of Waymo, the drivers have "safety" title.
Waymo is working on phasing out the safety drivers. But, what's the difference between that and Tesla's work on making your car achieve FSD?

The difference indeed is phasing out the drivers and the pace at which that is happening and the goal: car responsible driving, instead of driver responsible driving. It is a big, big difference.

Waymo has already offered driverless rides. Several manufacturers with MobilEye are aiming at limited scenario Level 3-4 systems in consumer cars for the next year and two, again systems where the driver does not need to be attentive. Sure these are still a work in progress today but they have a plan to start delivering Level 3+ rides soon and if Levandowski is to be believed these are happening sooner than his own approach is getting to that...

It will make all the difference in the world when you can sit in your car and read a book on the highway.

In the meanwhile, the other approach now is to develop increasingly good driver’s aids, but systems where the focus is not on handling the entire driving task, only a portion of it — which will thus continue to require an attentive driver. It is a very different focus, as Levandowski seems to readily admit too. It is made possible by an attentive driver.

 

[CarlK]

It will make all the difference in the world when you can sit in your car and read a book on the highway.

This I would say still falls in the ADAS territory. A true self driving car is one you could send to pick up kids at school. That is the most significant barrier that needs to be crossed. It's especially meaningful for companies like Uber, or Waymo that is after the Uber market. Cost of a "safety" personnel is the same as cost of a real driver if not more. You have gained nothing until you could eliminate need for that. On the other hand Tesla, or any car companies, could claim some degrees of success when its cars achieved the former.

 

[electronblue]

This I would say still falls in the ADAS territory. A true self driving car is one you could send to pick up kids at school. That is the most significant barrier that needs to be crossed. It's especially meaningful for companies like Uber, or Waymo that is after the Uber market. Cost of a "safety" personnel is the same as cost of a real driver if not more. You have gained nothing until you could eliminate need for that. On the other hand Tesla, or any car companies, could claim some degrees of success when its cars achieved the former.

Of course picking up the kids from school is the holy grail. But the daily commute will get a lot more productive or entertaining when stretches of it one can pass on the responsibility to the car (Level 3 and above).

 

[malcolm]

IIRC wasn't Blue Origin's New Shepherd the first reusable spacecraft to reach space? Gone a bit quiet since.

We've seen promising early announcements before, including sped-up curated videos.

Doesn't guarantee a robust solution is imminent.