17.7.2 - Why did Tesla release Local Autosteer vs...

[njxman]

With all the recent negative attention on safety and delays in releasing EAP functionality, I've been questioning the rationale at Tesla for choosing to release a dangerous and buggy version of Autosteer for local roads in Release 17.7.2 instead of these alternatives:
- Increasing TACC speed beyond 70 MPH
- Increasing Autosteer speed on Freeways/Interstates beyond 50 MPH
- Adding Autosteer on divided roads and highways (not just freeways)
- Adding Auto-wipers
- Adding Auto-headlights
- Adding Summons

The addition of Parallel Autopark in 17.7.2 was appreciated but personally, I would have preferred extending Autosteer on divided roads and highways well before local roads. Isn't going from Freeways -> Divided Highways -> Local roads seem to be to be the more obvious (and safer) path for extending Autosteer functionality?

 

[ClyffeWarden]

I suspect Tesla is trying to walk a fine line between keeping customers happy and avoiding lawsuits. Software bugs can be very hard to fix. I am hoping they can get it working sometime this year!.

 

[croman]

TACC is 80mph.

 

[davidc18]

It appears they underestimated the difficulty in starting over without mobile eye

 

[calisnow]

With all the recent negative attention on safety and delays in releasing EAP functionality, I've been questioning the rationale at Tesla for choosing to release a dangerous and buggy version of Autosteer for local roads in Release 17.7.2 instead of these alternatives:
- Increasing TACC speed beyond 70 MPH
- Increasing Autosteer speed on Freeways/Interstates beyond 50 MPH
- Adding Autosteer on divided roads and highways (not just freeways)
- Adding Auto-wipers
- Adding Auto-headlights
- Adding Summons

The addition of Parallel Autopark in 17.7.2 was appreciated but personally, I would have preferred extending Autosteer on divided roads and highways well before local roads. Isn't going from Freeways -> Divided Highways -> Local roads seem to be to be the more obvious (and safer) path for extending Autosteer functionality?

Neither you nor I work inside Tesla, do we?

 

[njxman]

Neither you nor I work inside Tesla, do we?

Do we need to be?!? I haven't seen a shortage of opinions (or criticism) on this forum from folks who I presume are not Tesla employees...

 

[BestRadar]

I finally got to test it today for 30 seconds and it tried to steer me into a curb. It was not ready for release.

 

[MP3Mike]

It appears they underestimated the difficulty in starting over without mobile eye

I don't think that is true, they knew it would take time which is why they planned for the initial AP2 cars to have both the Mobileye EyeQ3 chip and the NVidia DrivePX2 system. That way people could be using the AP1 portion while the AP2 portion was in shadow mode and Tesla got it up to AP1 level performance before they turned off the EyeQ3 chip.

But Mobileye wouldn't allow that and terminated their contract. So Tesla had to go all in on Tesla Vision.

I can sort of understand why Mobileye did that, they thought they could strong arm Tesla into giving up on Tesla Vision and just stick with their product. In the end all it did, for them, was reduce the number of EyeQ3 chips they sold and licensed their software for by ~100k. If I was a Mobileye shareholder I would be pissed that they refused that revenue stream.

 

[Shateam]

Tesla's approach to product releases (autopilot being the product here) is more like a technology company rather than a car manufacturer. Autopilot is a good example of a MVP (minimum viable product) wherein they don't wait for the product to be perfect before releasing it. The idea being, after the initial release, Tesla will continue to iterate, gain feedback and refine the product using an agile development methodology.

One can argue there are valid safety reasons to not use this approach but I think that argument has already been debated in (multiple) other threads...

 

[mrkisskiss]

It's easy to forget how experienced Mobileye are at this game. The vision processing they supplied for AP1 wasn't cutting edge - but it is very high quality; lest we forget that they've been doing it successfully for years and years.

The ImageNet training approach is still very new - even if Tesla aren't using Nvidia's DriveNet for AP2, they're probably still using similar techniques. Nvidia demoed some work in progress networks during their CES keynote last year, which they claimed they had trained a feature set that gave good object recognition in just two weeks. That idea that you can use deep learning techniques to very quickly train a driving specific computer vision feature set is relatively new...but it's just one piece of the puzzle.

Mobileye have trained their vision systems using manually annotated images - who knows how many hundreds of thousands... extremely labour intensive work. The Tesla/Nvidia approach is perhaps a shortcut, but perhaps not. Needless to say, they were definitely caught out with Mobileye tearing up their deal.

Here's what Musk actually said, taken from the 2016 Q4 earning's call (end of Feb 2017):

"Our original plan was to have a migration strategy where we have mobile eye and our Tesla Vision operating at the same time, to have kind of a smooth process, but Mobileye refused to do that so that forced us to respin the board, and caused unexpected delays, where we had to essentially delete their chip from the board and just {push through with} Tesla Vision. Safety is always our primary concern, and really we could have released Tesla Vision - including auto-steer at high-speed - like 3 months ago. I was driving it at high speed, personally, 3 months ago. But, we want to have an exhaustive testing process and validation process before enabling that across the fleet… so we’ve been edging our way up there gradually. Now longitudinal control or traffic aware cruise control is at 80mph, and autosteer is at 50mph, and I think we should be able to get - unless testing shows something different - we should be able to get them both to around 80-85mph next month, and be at parity with Hardware 1. And then, obviously things will only improve from there."

I pick up my AP2 car early April... Let's hope!

 

[Patrick W]

"Dangerous and buggy"? I must have gotten the safe and bug-free version. It's been working well for me. My only complaint its the speed restriction.

 

[NerdUno]

Just imagine you just bought a $140,000 condo only to find the indoor plumbing and electric service was still in beta and you'd need to use the porta-potty down the street for the next six months. If only you had read the builder's tweets.

 

[JasJ]

The auto steering is very wobbly and often suddenly gives up in AP2 (for me as of today). The car only seems to see cars directly in front and that seems to jump a lot.

I'm sticking with TACC for now as that works well up to 80.

I'm with others on adding the auto headlights, wipers, etc first while they keep tuning the steering and more cameras.

 

[EV-lutioin]

Just imagine you just bought a $140,000 condo only to find the indoor plumbing and electric service was still in beta and you'd need to use the porta-potty down the street for the next six months. If only you had read the builder's tweets.

I think Tesla made it pretty clear on their website configurator that the autopilot features would be rolled out over time. Your condo analogy only really works if you include that the buyer knew very clearly that certain features would be added later.

 

[182RG]

I think Tesla made it pretty clear on their website configurator that the autopilot features would be rolled out over time. Your condo analogy only really works if you include that the buyer knew very clearly that certain features would be added later.

No they didn't. Just stop.

Here's the configurator page for EAP as of Jan 2, 17. I'll pay you a bonus if you can find the words "rolled out over time".

Hint: you get $0

 

[DoubleACL]

Rolling out Summon, even if limited (such as forward and back 30') doesn't seem that difficult and I would use it more (like twice a day in.out of my garage than parallel park.

 

[S4WRXTTCS]

Yeah, it makes absolutely zero sense to me.

You come out with the low hanging fruit first. That means things like TACC, Summons, Auto-Park. All those don't rely heavily on what Mobile brought to the table.

Then you have FCW which doesn't have to be perfect.

Then Side/Blind spot indication with just the ultrasonic sensors at first.

Then some of the goodies like auto-wipers and auto-headlights.

They should have had all that by Jan 1st.

All those things people regularly use on a daily basis, or they have a chance of using every day.

People can live without lane-steering for awhile because it's doesn't even work that great in REAL rain. In most places outside of California they would have bought themselves at least 3-5 months to work on Lane-Steering. Owners would too busy complaining about auto-wipers to concentrate too much on lane-steering.

I'm sure even Tesla employees are scratching their heads wondering just what in the world they're doing.

 

[brianman]

... 17.7.2 ...

Some of us are still waiting to get seomthing newer than 17.5.x... :|

 

[182RG]

I'm sure even Tesla employees are scratching their heads wondering just what in the world they're doing.

They're covering themselves over their deceptive advertising. The one that uses the words "Enhanced Autopilot", and a first sentence that says "Enhanced Autopilot adds new capabilities to the Tesla Autopilot driving experience". Nope. They can't even get to AP1 parity.

People paid for it. They expected it by December 2016.

 

[EV-lutioin]

No they didn't. Just stop.

Here's the configurator page for EAP as of Jan 2, 17. I'll pay you a bonus if you can find the words "rolled out over time".

Hint: you get $0

View attachment 217264

Awesome, how much do I get!!