Tesla autopilot HW3

[CK_Stuggi]

What exactly did he get right this time around?
What DID they deliver?
I'm still sorely missing speed sign recognition in Europe, because the GPS data is crap (to put it nicely), only adjusting speed 100m behind the sign. While the cheaper competition (and AP1!) slow the car down before the limit comes into effect at the sign.
So, no delivery on AP1 parity. It took a whole year to get auto wipers. No refund for the downgrade.
Also, still no NoA in Europe for S/X, only 3 so far.

The overall quality of the software is appalling, compared to previous versions. In order to have proper functionality, a reset a day is in order.

What have they delivered? They have delivered a fishy redefinition of FSD features to cover less functionality and a video.
Essentially the whole Investor Autonomy event was a big explanation as to what they've been up to, and a video of a car going a preprogrammed round tour in easy traffic. That's actually less than they showed to have running, only needing "validation" in October 2016.

Do I hope they finally start delivering? Hell yeah!
But until I see something new, in my car, working reliably, I don't believe anything anymore.
And being on AP2, I'm in the group of the most effed over customers. At least I didn't lease, so I'll keep my car until they finally upgrade me, which they better do ASAP.

 

[CarlK]

What exactly did he get right this time around?

If you're paying attention to recent development Tesla's vision+deep learing is the right approach verses Lidat+HD map everyone else is doing. Even the original Google/Waymo Lidar guru now says "Elon is right". Tesla's been taking this route alone for years. If you have problem with Tesla I can assure you that you will not get anything from anyone else in the foreseeable future that you will like.

The rest of your whine are not unrelated to the thread subject which is HW3. Just want to point that out.

 

[electronblue]

But until I see something new, in my car, working reliably, I don't believe anything anymore.
And being on AP2, I'm in the group of the most effed over customers. At least I didn't lease, so I'll keep my car until they finally upgrade me, which they better do ASAP.

My thoughts exactly.

Anyway I guess all this boils down to wait and see. How right or wrong Elon was on April 22nd and how right or wrong the competition is. I don’t see how we’ll be any the wiser until next year so see you then.

For my car’s sake I really wish Elon finally was truthful this time around.

 

[J1mbo]

m still sorely missing speed sign recognition in Europe, because the GPS data is crap

Not where I live. AP2.0 worked really well. Reliable, too.

Currently driving an i3 with speed sign recognition while I wait for my M3 to arrive.

Suggest you try the i3's sign recognition system - it will redefine your definition of "crap".

 

[rnortman]

According to Elon at this point in time.
In October 2016, feature parity was to come end of December 2016. Didn't happen.
In March 2017, FSD features were 3 months maybe, 6 months definitely. Didn't happen.
End of 2017, coast to coast was a few months away. Didn't happen.
Now it's feature complete FSD end of 2019.

You forgot "In summer 2018, first FSD features coming in August".

 

[rnortman]

If you're paying attention to recent development Tesla's vision+deep learing is the right approach verses Lidat+HD map everyone else is doing. Even the original Google/Waymo Lidar guru now says "Elon is right". Tesla's been taking this route alone for years. If you have problem with Tesla I can assure you that you will not get anything from anyone else in the foreseeable future that you will like.

You do know that EVERYBODY developing autonomous vehicles uses cameras and deep learning? Literally every company doing this has vision systems like Tesla's, except generally with better cameras and more powerful inference hardware. And I'm sorry, which "guru" says "Elon is right"? And about what exactly?

 

[diplomat33]

It doesn't say "all functions of automatic city driving" it doesn't say "feature complete auto city driving"

So I'd bet that it does not include 90 degree turns... or at least that is not on the road map for 2019.


Right now NoA does "Automatic highway driving" .... but it doesn't do several types of double merges, cloverleafs, pulling over for emergency vehicles, filter lights entering highways, and many other things.


And I know that auto lane changes has not been implemented on non- highways..... I was asking... if it was added... why is there any more of a concern of auto lane changes on city streets than highways?

Furthermore, I'm not even convinced that auto lane changes is included in "Automatic driving on city streets". or at least is not on their roadmap for what's coming this year. But it could be I'm not sure.

I imagine 90 degree right turns would be added first. But yes, complex 4 lane intersections and roundabouts will probably be added later.

The reason I think auto lane changes on city streets are important is because there will be instances where you need to do switch lanes for example, to get into a right turn only lane, in order to follow your navigation directions. So moving over to a right or left turn only lane before an intersection will be a very basic feature that needs to be implemented early in order to do "automatic city driving".

 

[turnem]

I imagine 90 degree right turns would be added first. But yes, complex 4 lane intersections and roundabouts will probably be added later.

The reason I think auto lane changes on city streets are important is because there will be instances where you need to do switch lanes for example, to get into a right turn only lane, in order to follow your navigation directions. So moving over to a right or left turn only lane before an intersection will be a very basic feature that needs to be implemented early in order to do "automatic city driving".

I hope you're right. I would be blown away if they deliver this on the front end of the city street FSD rather than the back end.

I'm hopeful but somewhat pessimistic because I'm still seeing lots of instances that AP needs to improve while on city streets. I'm still fairly new and I'm impressed by the technology but it still has a long way to go. There are multiple instances where AP really struggles.

• One lane splitting into two. It seems to favor the left side so if you're in the right lane it will try to abruptly change to the left lane.
• Intersections with slight curves. The car struggles sometimes (but at other times does amazingly well) to understand which lane it should be heading into. Same as above - it strongly favors the left. If you allow it, the car will want to sometimes abruptly switch to the left lane
• Sharper right hand turns - it seems to turn too late which results in hugging (or even sometimes slightly straddling) the left hand lane marker.
• Recognizing stopped vehicles that are around a slightly turn (sharp turns excluded - these are hard for humans too). The car WILL stop but it is likely to require a change of underwear.

Don't get me wrong. I'm a huge fan. The technology is super impressive. And there's a much longer list of things it does well than things it doesn't do so well. But my point is that many of the scenarios above are fairly critical to get sorted out before FSD works reliably (or even acceptably well) on city streets.

 

[turnem]

Two things that I think could help greatly with AP:

1. Reading road signs. Like reading a sharp turn sign and knowing to slow down a little.
2. Reading blinkers on cars in front of you. This would help with the odd TAAC behavior when a car turns out of your lane but the car almost comes to a complete stop and then resumes very slowly/cautiously.

Does anyone know if they have plans to go in this direction?

 

[electronblue]

Not where I live. AP2.0 worked really well. Reliable, too.

Currently driving an i3 with speed sign recognition while I wait for my M3 to arrive.

Suggest you try the i3's sign recognition system - it will redefine your definition of "crap".

Except AP2.0 does not recognize traffic signs at all yet. It is all database.

 

[diplomat33]

I hope you're right. I would be blown away if they deliver this on the front end of the city street FSD rather than the back end.

I'm hopeful but somewhat pessimistic because I'm still seeing lots of instances that AP needs to improve while on city streets. I'm still fairly new and I'm impressed by the technology but it still has a long way to go. There are multiple instances where AP really struggles.

• One lane splitting into two. It seems to favor the left side so if you're in the right lane it will try to abruptly change to the left lane.
• Intersections with slight curves. The car struggles sometimes (but at other times does amazingly well) to understand which lane it should be heading into. Same as above - it strongly favors the left. If you allow it, the car will want to sometimes abruptly switch to the left lane
• Sharper right hand turns - it seems to turn too late which results in hugging (or even sometimes slightly straddling) the left hand lane marker.
• Recognizing stopped vehicles that are around a slightly turn (sharp turns excluded - these are hard for humans too). The car WILL stop but it is likely to require a change of underwear.

Don't get me wrong. I'm a huge fan. The technology is super impressive. And there's a much longer list of things it does well than things it doesn't do so well. But my point is that many of the scenarios above are fairly critical to get sorted out before FSD works reliably (or even acceptably well) on city streets.

I totally get people's skepticism about FSD since there is a lot left for Tesla to do before the system is a reliable city level self-driving system. Heck, Tesla started with highway driving, which is the easy part, and there is still some things left to do there. City self-driving is even harder.

It is worth noting though that the AP that we are using now is not designed for city driving. So all the shortcomings you mention are not a true indicator of what Tesla can do in terms of city driving. When we get "automatic city driving" on AP3, then we can measure how close Tesla is at city self-driving or not.

 

[turnem]

I totally get people's skepticism about FSD since there is a lot left for Tesla to do before the system is a reliable city level self-driving system. Heck, Tesla started with highway driving, which is the easy part, and there is still some things left to do there. City self-driving is even harder.

It is worth noting though that the AP that we are using now is not designed for city driving. So all the shortcomings you mention are not a true indicator of what Tesla can do in terms of city driving. When we get "automatic city driving" on AP3, then we can measure how close Tesla is at city self-driving or not.

Yep. We'll just have to wait and see I suppose. I would like to see them start rolling out some of these more advanced capabilities even in advance of rolling out true FSD.

Having said that - AP has truly come a long way. I'm a believer and I'm having fun being a part of the ride.

 

[diplomat33]

Yep. We'll just have to wait and see I suppose. I would like to see them start rolling out some of these more advanced capabilities even in advance of rolling out true FSD.

Having said that - AP has truly come a long way. I'm a believer and I'm having fun being a part of the ride.

Oh yes. I am a believer too. I've seen the progress of AP2. And at the end of the day, the things that Tesla needs to accomplish to achieve FSD are not out of reach. And Tesla can add features piece by piece via software update. it will be great to watch.

 

[Maitri982]

What exactly did he get right this time around?
What DID they deliver?
I'm still sorely missing speed sign recognition in Europe, because the GPS data is crap (to put it nicely), only adjusting speed 100m behind the sign. While the cheaper competition (and AP1!) slow the car down before the limit comes into effect at the sign.
So, no delivery on AP1 parity. It took a whole year to get auto wipers. No refund for the downgrade.
Also, still no NoA in Europe for S/X, only 3 so far.

The overall quality of the software is appalling, compared to previous versions. In order to have proper functionality, a reset a day is in order.

What have they delivered? They have delivered a fishy redefinition of FSD features to cover less functionality and a video.
Essentially the whole Investor Autonomy event was a big explanation as to what they've been up to, and a video of a car going a preprogrammed round tour in easy traffic. That's actually less than they showed to have running, only needing "validation" in October 2016.

Do I hope they finally start delivering? Hell yeah!
But until I see something new, in my car, working reliably, I don't believe anything anymore.
And being on AP2, I'm in the group of the most effed over customers. At least I didn't lease, so I'll keep my car until they finally upgrade me, which they better do ASAP.

Can't argue with much of this except software quality being "appalling". He's basically been full of *sugar* on FSD and related items for some time now. Hard to have anyone believe a word out of your mouth when that's your track record.

I am a big fan of his in many ways, but in others he seems to lack some basic common sense as to when to STFU and just keep your head down and deliver the goods to the shock of the world when and only when you have them beyond a doubt.

 

[Maitri982]

...

• Recognizing stopped vehicles that are around a slightly turn (sharp turns excluded - these are hard for humans too). The car WILL stop but it is likely to require a change of underwear.

.

I have seen this too and I don't have the kahunas to see if it will stop or plow into a bunch of cars killing people. So it that to the heap of disengagements of NOA from me.

 

[CarlK]

You do know that EVERYBODY developing autonomous vehicles uses cameras and deep learning? Literally every company doing this has vision systems like Tesla's, except generally with better cameras and more powerful inference hardware.

Either you have confused vision primary system to Lidar primary system with camera or you just want to add confusion to others. They are totally different systems. Watch the Lex Fridman video (41:48). NOBODY other than Tesla is using vision+deep learning for the simple reason that nobody has a fleet large enough to that to do vision + deep learning. That's especially impossible if you want to put a Lidar in every car.

And I'm sorry, which "guru" says "Elon is right"? And about what exactly?

About the Lidar system. Levandowski was the original Google Lidar guru and responsible for the entire system back from the begining. He said in a recent interview appropriately titled lessons learnd that "a wiser person than me said Lidar is a crutch" and "Elon is right" (go to 1:37). This and the fact that there was zero Lidar guy even came out to argue what Elon and Kaparthy said pretty much settled this whole thing.

Everything everyone one, including many on this forum, said about FSD was wrong. Of course those do not matter (they really do not) we just need to continue to attack Elon which was the only person that was right.

 

[mongo]

Here is today's nth 9 edge case: A power company truck dragging two utility poles down the road being followed by a pickup truck with a guy hanging out the passenger side window...

 

[CarlK]

@turnem Most of your concerns are valid but they were pretty much all answered in the investor event. The deep learning is not only powerful enought to recognize speed signs, stop lights etc but could even drive on snow covered roads and making right predition of curves we can not see past (human drivers actually have this capability). The most significant one we first heard of is the Tesla system can observe small movement of cars and pedestrians to predict their actions (changine lanes or walk off the curb). Again it is using the fleet learning to do it in the shadow mode even if the AP is not turned on. The AI will just make preditions and then compare them to what actually happen to train and fine tune the machine. It's super smart and powerful approach and probably solved the last missing link of self driving cars. That's what Levandowski aluded to in his short interview I linked above.

 

[CarlK]

Who is that smart guy who disagree with Lex Fridman and Levandowski and without even bother to give the arument of why? @rnortman you have just showed your true color as another worthless troll. Sorry that I wasted my time on you. Off you go to my ignore list too.

 

[CK_Stuggi]

If you're paying attention to recent development Tesla's vision+deep learing is the right approach verses Lidat+HD map everyone else is doing. Even the original Google/Waymo Lidar guru now says "Elon is right". Tesla's been taking this route alone for years.

That is not quite correct. There's a very interesting video from Mobileye demonstrating a camera-only (NO LIDAR) vehicle merging "assertively" into traffic.
The video is from May 2018, citing implementation with major automakers in July (no idea if that happened, if it's providing the buyers with ADAS or is just helping data collection)(relevant section starts at 30 min in)

And, yes, I've followed the development and the sensor approach with keen interest. And I really hope all the claims eventually come true. It's a fun ride (mostly) to see features delivered. But the track record of announcement is a desaster and once you bump your car into a wall because the most recent update broke front ultrasonics, the fun stops (happened about a year ago, to very few users, me among them).
Hence, AP/FSD announcements by Tesla are like electric car announcements by German automakers - looks good in PowerPoint, I'll believe it when it's delivered.