Anthony Levandowski beats Tesla's Elon Musk to first Auto Cross country (3k Miles, 0 disengagement)

[diplomat33]

I'm sorry but until Elon's NOA can do a cross country drive with 3,099 consecutive miles with no interventions, without touching the wheel or steering wheel. Then he's behind co-pilot.

Except that's not at all what the video showed! The video showed a car go a few hundred miles with no driver input, then the driver disengaged to pull over for gas or to sleep, then it got back on the highway, did a few more hundred miles with no driver input. rince repeat until he did 3,099 miles.

 

[Tam]

...literally meanless.

Waymo can't even autonomously get their cars out of their garage without a human turning the steering wheel for a left turn.



Waymo hasn't demonstrated that it can find a random parking space just yet.

 

[ucmndd]

I'm sorry but until Elon's NOA can do a cross country drive with 3,099 consecutive miles with no interventions, without touching the wheel or steering wheel. Then he's behind co-pilot.

I certainly didn’t see 3,099 consecutive miles without disengagement on that video.

“No interventions (except for the hard parts)” is silly.

 

[ucmndd]

It your issue with Tesla or with people who, you feel, think too much of Tesla?

I’d presume both. OP’s entire reason for participating on this forum appears to be to protect and defend the interests of their employer.

 

[diplomat33]

@Bladerskb You are misrepresenting NOA and Levandowski's system. You are painting NOA as worst than it really is. I use it and while it is certainly far from perfect, it is not "crap". And you are painting his system as much better than it really is. The video shows a system that went without driver input for a few hundred miles, then the driver had to take control to pull over for gas or to sleep, then the driver gets back on the highway and goes a few more hundred miles with no driver input. rince repeat for 3000 miles. And yet, you are acting like it is some amazing self-driving system that is 100x better than AP, which it clearly is not.

 

[Bladerskb]

Trying to resist urge to feed the trolls, but:

You are (purposely) conflating two very different takes on what constitutes a “cross country drive”. Tesla’s stated goal for the cross-country drive was basically a long-distance summon - you call the car from across the country, it finds you at the other end. Charging stops, everything. All of the on/off the interstate complexity involved.

If you fail to acknowledge the difference between that and what was demonstrated by Levandowski (extended interstate cruising with manual intervention by a driver every 500 miles to handle all of the complex tasks like navigating to a fuel station or rest stop), you’re only deluding yourself.

Long distance summon? No because a long distance summon as stated by elon is level 5 car.

However the cross country that was set for end of 2017 and delayed every month afterwards was supposed to be what co-pilot just did plus under 10 miles of city street driving. How is it that you dont understand. A general city street level 2 system that has a reliability of 50 miles before disengagement would be able to do it.

It doesnt constitute a difference. It's literally the same thing but 10 miles of city driving.

Except that's not at all what the video showed! The video showed a car go a few hundred miles with no driver input, then the driver disengaged to pull over for gas or to sleep, then it got back on the highway, did a few more hundred miles with no driver input. rince repeat until he did 3,099 miles.

3,099 consecutive highway miles. NOA fails in like 10 miles give or take.

How about this. When no confirmation is released next month. You replicate that route from the very start and see how far you can go. You wouldn't last 100 miles period.

And if you read between the lines, FSD isn't where he's going. He's going for long haul truckers that run on Interstates, barely a notch above Cadillacs capabilities.

So supercruise can now autonomously change lanes and handle highway interchanges?

 

[mongo]

Long distance summon? No because a long distance summon as stated by elon is level 5 car.

Just because he didn't add a disclaimer regarding the environmental conditions to whatever statement you are referencing, does not mean Elon meant level 5 vs level 4.

 

[diplomat33]

NOA fails in like 10 miles give or take.

Again, that's just patently not true. I've done more than 10 miles on NOA easily without any disengagement.

How about this. When no confirmation is released next month. You replicate that route from the very start and see how far you can go. You wouldn't last 100 miles period.

Well, I've done over 100 miles on AP with no disengagement. So I've already done it.

 

[Russell]

3,099 consecutive highway miles.

No it wasn't. Levandowski blacked out the video when he took exit ramps to refuel and to sleep.

 

[Bladerskb]

However, when the Guardian is there for a 48-mile test ride:

...at one point, Levandowski took the wheel after the car failed to merge into busy traffic. Such hiccups are called disengagements in the self-driving world. Levandowski attributed the disengagement to the latest version of Pronto’s constantly evolving software."

That was on SF city streets. The author of the guardian simply didn't clarify.

That's total BS. First, he had disengagements, he just hid them in the promo video, and second, the car failed to merge into busy traffic so it is not better than NOA.

Huh the cross country highway drive had no disengagements. The drive that had disengagement was with The Guardian and that was on SF city streets.

 

[morrisdl]

I dont see how this is much better than my AP1 S that drove 5,415 of the 5730 (>95%) in our summer road trip. I can count on one hand the number of disengagements that were not directly for rests or refuels stops.

 

[diplomat33]

Huh the cross country highway drive had no disengagements. The drive that had disengagement was with The Guardian and that was on SF city streets.

Yes the cross country drive did have disengagements because he had to disengage every time he left the highway to get gas or stop to sleep. So it was not a true cross country self driving demo.

 

[Bladerskb]

Yes the cross country drive did have disengagements because he had to disengage every time he left the highway to get gas or stop to sleep. So it was not a true cross country self driving demo.

So you were proven wrong about your merge into traffic comment, now you are on to another.


That's like saying a Level 4 highway system had a failure disengagement because the driver was ready to leave the highway...like really? You tesla fans are something else.

 

[Chaserr]

I certainly didn’t see 3,099 consecutive miles without disengagement on that video.

“No interventions (except for the hard parts)” is silly.

I was hoping to see fuel stops automated. It looks like they're just running openpilot with nags disabled. There were no off-highway to refueling stops shown, which makes me think it's just a lanekeeper they used for a few thousand miles. Coast to coast without disengagements means the whole drive, not some portions of it and please ignore the disengaged parts with traffic signs and lights.

 

[S4WRXTTCS]

As a tech demo I liked it, but once again a tech demo presents what they want to show. You can't tell a whole lot because you can't slow down the video slow enough, and with enough detail to see how it handled various situations.

Sure it's neat, and I saw some neat mode switching that I thought was cool. Like when it would switch to hazard driving or construction driving.

But, at the end of the day it's still a tech demo that was done in good weather conditions. I'm really surprised they were able to do a cross country trip in Late October without encountering any significant rain fall or low visibility.

I'm not sure how Tesla/Elon plays into this other than it's a good display of what NOA should behave like as.

I already give NOA a hard time (or an D- rating) because it feels like it's more for showing off than something that works. I tried it from north of Seattle to Portland, and it was absolute garbage.

But, the thing is if I programmed the system I could make it work correctly by taking into account what to expect.

The biggest issue was maps, and that it simply didn't have the info it needed. It didn't even know what lane it needed to be in from 405S to I5S. Then it tried to convince me to get it into the left lane on I5S because supposedly my lane was going to end, but it wasn't ever going to end.

The lane change recommendations for overtaking were also pointless without any understanding of the rules of the road or how traffic was flowing. Eventually I turned off the lane change recommendations for passing. Note: In the latest version (which I don't have yet) it does seem like they fixed this issue.

In WA state it was frustrating to use NOA in the right lane because any time there was a merging on ramp it tried to recenter so it must have looked completely ridiculous to anyone behind me. Like what is that idiot doing? It didn't happen in OR because they use dashed lines on the right for the merging on ramp. In the middle lane it was a great, and rock steady.

 

[Bladerskb]

I was hoping to see fuel stops automated. It looks like they're just running openpilot with nags disabled. There were no off-highway to refueling stops shown, which makes me think it's just a lanekeeper they used for a few thousand miles. Coast to coast without disengagements means the whole drive, not some portions of it and please ignore the disengaged parts with traffic signs and lights.

Wait openpilot can do lane changes and handle highway interchanges? Gotta love the hot takes here. Tesla fans never cease to amaze me.

 

[Chaserr]

Yeah it's getting better than AP1 lately. More developers working on it, so it was inevitably going to pass AP1. They'll need more camera support to catch AP2.

 

[Vitold]

It is a safe bet that as more information comes out we will be less and less impressive.
Overall Pronto system demonstrated that it is somewhat capable of driving on highways (I read that it took three tries).

 

[Bladerskb]

Yeah it's getting better than AP1 lately. More developers working on it, so it was inevitably going to pass AP1. They'll need more camera support to catch AP2.

So it CAN'T do autonomous lane changes nor handle highway interchanges. Gotcha.

 

[Bladerskb]

@Bladerskb You are misrepresenting NOA and Levandowski's system. You are painting NOA as worst than it really is. I use it and while it is certainly far from perfect, it is not "crap". And you are painting his system as much better than it really is. The video shows a system that went without driver input for a few hundred miles, then the driver had to take control to pull over for gas or to sleep, then the driver gets back on the highway and goes a few more hundred miles with no driver input. rince repeat for 3000 miles. And yet, you are acting like it is some amazing self-driving system that is 100x better than AP, which it clearly is not.

If it takes NOA 100 disengagement to do a cross country drive then Co-Pilot IS 100x better than NOA.

Don't take my word for it. Take tesla owners word for it.

Non-Confirm Lane change next month : teslamotors
Navigate on Autopilot is Useless (2018.42.3)

NOA was supposed to prove that Tesla is 3+ years ahead of the competition.
Co-pilot is made by one person with zero funding and is 100x better than NOA.

If Co-pilot is better than NOA then how is Tesla far ahead?