Levandowski drives coast to coast without touching steering

[mspohr]

Controversial engineer: I travelled over 3,000 miles in a self-driving car
Controversial engineer: I travelled over 3,000 miles in a self-driving car

The car, a modified Toyota Prius, used only video cameras, computers and basic digital maps to make the cross-country trip.

Levandowski told the Guardian that, although he was sitting in the driver’s seat the entire time, he did not touch the steering wheels or pedals, aside from planned stops to rest and re-fuel. “If there was nobody in the car, it would have worked,” he said.
The system does not use laser-ranging lidars like those that Levandowski helped to develop at Waymo, Otto and Uber. This is not because he is afraid of more lawsuits, Levandowski insists, but because he now believes that lidars are an expensive and unnecessary red herring in the quest for robotic vehicles.

 

[Bladerskb]

WITH NO DISENGAGEMENT!

Some of the comments I posted earlier was...

Delphi/Aptiv did 99% in 2015 and now Anthony Levandowski with his new company Pronto.AI has done 100% in 2018 using just cameras and regular navigation maps (no pre-mapping or HD mapping).

This drive by Anthony shows you where the state of the art is by one guy with a very small team with no funding. This is Another blow to Tesla, Elon and the fanbase. Shows you how easy it is to do a cross country yet Tesla are still struggling. Alot of people like @strangecosmos look at their release of AP and think Tesla is 3, 5, 10 years ahead because of NOA. but this shows you that the state of the art is well beyond what AP is. NOA actually reveals how behind Tesla is.

but because he now believes that lidars are an expensive and unnecessary red herring in the quest for robotic vehicles.

He says this now because he doesn't have any funding to include Lidar for redundancy.

 

[Unpilot]

WITH NO DISENGAGEMENT!

Some of the comments I posted earlier was...

Delphi/Aptiv did 99% in 2015 and now Anthony Levandowski with his new company Pronto.AI has done 100% in 2018 using just cameras and regular navigation maps (no pre-mapping or HD mapping).

This drive by Anthony shows you where the state of the art is by one guy with a very small team with no funding. This is Another blow to Tesla, Elon and the fanbase. Shows you how easy it is to do a cross country yet Tesla are still struggling. Alot of people like @strangecosmos look at their release of AP and think Tesla is 3, 5, 10 years ahead because of NOA. but this shows you that the state of the art is well beyond what AP is. NOA actually reveals how behind Tesla is.



He says this now because he doesn't have any funding to include Lidar for redundancy.

You don't seem to like Elon or Tesla very much. Yet you have over 1k post's on a Tesla fan forum.

Keep on fighting the "good" fight...I guess

 

[Bladerskb]

You don't seem to like Elon or Tesla very much. Yet you have over 1k post's on a Tesla fan forum.

Keep on fighting the "good" fight...I guess

I dont hate anyone. I simply navigate the path as it unfolds

 

[Bladerskb]

Pronto Means Ready – Pronto AI – Medium


3 2 1.. before @strangecosmos claims Anthony's use of only cameras and focus on imitation learning proves that Tesla is on the right track and will achieve level 5 in 2019.

 

[Bballou]

Levandowski told the Guardian that, although he was sitting in the driver’s seat the entire time, he did not touch the steering wheels or pedals, aside from planned stops to rest and re-fuel.

That's not 100%. Tesla's intention is 100% coast-to-coast, including stops for recharging.

Elon stated in the Q2 earnings call:

We could do a coast-to-coast drive, especially if we pick a specific route and write code to make it work, but that would be kind of gaming the system.

If anything, this is just further evidence that Tesla's no-LIDAR approach is viable. In my opinion, Tesla is doing the right thing by focusing on the general solution to self-driving, rather than special cases for theatrical effect. When Tesla is able to do a truly 100% coast-to-coast drive, it will be far more impressive because it means the entire fleet is close to that capability.

 

[Bladerskb]

That's not 100%. Tesla's intention is 100% coast-to-coast, including stops for recharging.

Elon stated in the Q2 earnings call:

Cross country drive involves 3k miles of highway driving and only includes 10-20 miles of city street driving.

So yes they did do 100% cross country. Because cross country is about highways.

 

[Bballou]

Cross country drive involves 3k miles of highway driving and only includes 10-20 miles of city street driving.

So yes they did do 100% cross country. Because cross country is about highways.

That's like saying you can 100% fly a plane, even if you can't take off or land it.

 

[mspohr]

That's not 100%. Tesla's intention is 100% coast-to-coast, including stops for recharging.

Elon stated in the Q2 earnings call:


If anything, this is just further evidence that Tesla's no-LIDAR approach is viable. In my opinion, Tesla is doing the right thing by focusing on the general solution to self-driving, rather than special cases for theatrical effect. When Tesla is able to do a truly 100% coast-to-coast drive, it will be far more impressive because it means the entire fleet is close to that capability.

It looks like Levandowski uses a neural net, just like Tesla. He uses six low resolution cameras and no LIDAR. It does validate Tesla's approach.
This was clearly a publicity stunt but it worked and gives us all hope that Tesla's fleet will someday (hopefully soon) be capable of autonomous driving. If Levandowski can do this with very limited resources, I think Tesla can do it also.

 

[acoste]

It looks like Levandowski uses a neural net, just like Tesla. He uses six low resolution cameras and no LIDAR. It does validate Tesla's approach.
This was clearly a publicity stunt but it worked and gives us all hope that Tesla's fleet will someday (hopefully soon) be capable of autonomous driving. If Levandowski can do this with very limited resources, I think Tesla can do it also.

I do no think this validates anything regarding Lidar. Nice achievement for sure, but how many corner cases did he see on the road? 1 or 2?

Cameras only solution has no redundancy and as I mentioned earlier this will be the poor man's autopilot due to its marginal safety.

Responsible car manufacturers are not releasing cars without redundancy. Audi's latest partnership with Lumentum confirms this.

 

[mspohr]

I do no think this validates anything regarding Lidar. Nice achievement for sure, but how many corner cases did he see on the road? 1 or 2?

Cameras only solution has no redundancy and as I mentioned earlier this will be the poor man's autopilot due to its marginal safety.

Responsible car manufacturers are not releasing cars without redundancy. Audi's latest partnership with Lumentum confirms this.

Not sure I understand what you are saying...
He didn't use LIDAR so of course it proves nothing about LIDAR.
He used cameras and a neural net so it shows that it worked for 3000 miles... an impressive achievement.
I don't understand your point about redundancy. It uses six cameras... that seems redundant. A single LIDAR would provide a seventh view but is this necessary? Clearly not since he made it 3000 miles with only six cameras.

 

[acoste]

Not sure I understand what you are saying...
He didn't use LIDAR so of course it proves nothing about LIDAR.
He used cameras and a neural net so it shows that it worked for 3000 miles... an impressive achievement.
I don't understand your point about redundancy. It uses six cameras... that seems redundant. A single LIDAR would provide a seventh view but is this necessary? Clearly not since he made it 3000 miles with only six cameras.

The trip does not prove that camera only solution is the way to go.

Redundancy may mean 2 cameras doing the same thing simultaneously but the best way is if it is based on different technologies. Cameras may miss to recognize an object even if it's a big freaking truck as the many Tesla accidents show it.

 

[electronblue]

Great job! Interesting stuff.

 

[Bballou]

I do no think this validates anything regarding Lidar. Nice achievement for sure, but how many corner cases did he see on the road? 1 or 2?

I'm not sure how LIDAR helps a vehicle handle corner cases. It seems like intelligent image processing would detect a far broader range of unusual events than LIDAR. The ability to identify and predict corner cases seems like a question of neural network capability, with or without LIDAR.

 

[acoste]

I'm not sure how LIDAR helps a vehicle handle corner cases. It seems like intelligent image processing would detect a far broader range of unusual events than LIDAR. The ability to identify and predict corner cases seems like a question of neural network capability, with or without LIDAR.

First thing is sensing. Even if we assume an intelligent image processing, cameras are not very good at night or in rain or if the windshield gets dirty. Or facing the sun on a wet road. Worst, the combination of these. Second if some calibration is off, distance measurement using a different technology can confirm it.

 

[electronblue]

I'm not sure how LIDAR helps a vehicle handle corner cases. It seems like intelligent image processing would detect a far broader range of unusual events than LIDAR. The ability to identify and predict corner cases seems like a question of neural network capability, with or without LIDAR.

The thing that separates Lidar from camera is practically no false negatives. Camera must identify to act, but when Lidar says nothing is there, nothing is there at a certain distance. Lidar is very accurate in saying nothing is in front of the car for example.

This is why Uber failed. Their vision failed to recognize the oddly shaped shopping cart filled with stuff in the middle of the road and their Lidar was disabled. Lidar would have known with very high reliability that something is there. Plus it has none of the negatives of radar and ultrasonics that miss certain types of targets or exaggerate them.

This is what makes Lidar useful for redundancy. There are other benefits but this just from a redundancy point of view.

 

[Engr]

WITH NO DISENGAGEMENT!

Some of the comments I posted earlier was...

Delphi/Aptiv did 99% in 2015 and now Anthony Levandowski with his new company Pronto.AI has done 100% in 2018 using just cameras and regular navigation maps (no pre-mapping or HD mapping).

This drive by Anthony shows you where the state of the art is by one guy with a very small team with no funding. ..... blow to Tesla...

He says this now because he doesn't have any funding to include Lidar for redundancy.

> "State of the art.."
Do you really believe that? I forgot you're always correct.

> "because he doesn't have any funding to include Lidar..."

In reality, this is what he said:
"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."


After his his dig lidar, he also said:
"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." I wonder what are these "crutch" technologies.!?.

Anyway, @Bladerskb said this is the state of the art.

Side note, according to state of the art:
"Deespite the vast sums of money and time dedicated to developing and rolling out autonomous vehicles, there are no real autonomous vehicles today. There are only increasingly complex and expensive demonstrations of those visions I mentioned above.

Second, true level 4 or 5 vehicles will not arrive for many more years."

Edit to add Levandowsky's post here:Pronto Means Ready – Pronto AI – Medium

 

[scaesare]

This is Another blow to Tesla, Elon and the fanbase. Shows you how easy it is to do a cross country yet Tesla are still struggling.

I KNOW!!

It's just like the U.S. and Russia in the space race all over again... there's NO WAY the U.S. is going to catch up the Soviets and beat them to the moon!


Oh... Wait.

 

[electronblue]

@Engr That of course is the Elon Musk, George Hotz — should I say startup Silicon Valley — view. More of the same. The thing is, can they achieve real legal car responsible driving (no driver responsible) with cameras only and how fast?

Car responsible driving is not a false dream. When you can read a book on the highway that will be a big deal.

Also, why disparage technology that provides redundancy. Redundancy can well mean better safety.

 

[Bballou]

First thing is sensing. Even if we assume an intelligent image processing, cameras are not very good at night or in rain or if the windshield gets dirty. Or facing the sun on a wet road. Worst, the combination of these. Second if some calibration is off, distance measurement using a different technology can confirm it.

You aren't going to have a self-driving car without cameras. If the cameras can't see, then it's not safe for anyone or anything to drive, including LIDAR equipped cars.

Self-driving cars with superhuman capabilities would be great and I agree that LIDAR doesn't hurt the capabilities of a car, but I still don't see how it's necessary to achieve a human-parity self-driving car. Self-driving cars can wait for safe driving conditions, just like people.

Distance measurements are sufficiently confirmed by the far less expensive radar in the places where it matters. Even here though, I think current iterations of AP and other solutions rely too heavily on radar, resulting in false-positives causing sudden braking or data suppression of actual obstacles. The most complete solution to these problems is more use of cameras and better image processing.

This is why Uber failed. Their vision failed to recognize the oddly shaped shopping cart filled with stuff in the middle of the road and their Lidar was disabled. Lidar would have known with very high reliability that something is there. Plus it has none of the negatives of radar and ultrasonics that miss certain types of targets or exaggerate them.

I assume you mean bicycle, which was behind the person, so isn't really relevant (what if the person wasn't pushing anything?). I've attached a screenshot from the video.

Disregarding the culpability of the safety-driver, to me this collision demonstrates:
1) an insufficiently capable neural net
2) possibly insufficient optical sensors (there really isn't anyway to determine the capabilities of the optical sensors based on this video. It may be far more capable at night that what is represented by the video)
3) vehicle travelling faster than its detection capabilities permitted (a programming problem)

The UBER vehicle should have been using LIDAR, but only because it's remaining sensors and image processing capabilities were inadequate for the speed it was going. If inexpensive self-driving cars have to travel slower at night, so be it. It's still possible without LIDAR.