Does AP2 mean the end of AP1 SW dev?

[Tslacg]

I've gotta say, the split between mobil eye and Tesla seemed a bit sudden. The rollout of AP2 and some of the less than sleek AP2 camera integration makes me think the transition was rushed, and there was less remaining AP1 hardware Tesla had in Stock to put on new vehicles than would have been desired. This would also mean the software development for AP2 wasn't ready, thus the delayed rollout and need for field verification. I know this happened with AP1, but then it was more critical to release the 'next big thing'. This release for AP2 seemed forced.

All of this is fine, but does worry me that the recent 8.0 changes to Apt for my car are here to stay. I'd be very frustrated that my model s that used to handle sunny days on the highway just fine, and had finally quit lurching to the right over hills on (7.1) will now be stuck with this behavior in 8.0 & 8.1.

What does everyone think? Are the refinements to AP1 finished, and will the AP1 SW team be focusing all efforts on AP2?

 

[sdorn]

The AP1 cars all use the Mobileye hardware and software. From what we've been told the AP2 cars use nVidia hardware and software developed in-house by Tesla. The articles that reported on the Tesla / Mobileye split indicated that Mobileye would continue to support ongoing upgrades to the AP1 software, but given the supply contract has been terminated it seems unlikely they will be working that hard on software upgrades for AP1 hardware. It seems more likely it will just be maintenance upgrades and bug fixes.

 

[kdday]

My conspiracy theory is not that AP 2.0 hardware was rushed, but rather the accompanying software features were deliberately held back in order to incentivize buyers to still buy remaining AP 1.0 inventory cars.

Much easier to convince a buyer to buy an AP 1.0 car with features the 2.0 cars still don't have.

 

[AWDtsla]

AP 1 is basically dead. I imagine more restrictions may be in the pipeline, but I expect no improvement in any functionality. If nothing else maybe stop the car from veering off the road because hundreds of Tesla's before you merged abruptly without blinkers.

 

[andrewket]

My conspiracy theory is not that AP 2.0 hardware was rushed, but rather the accompanying software features were deliberately held back in order to incentivize buyers to still buy remaining AP 1.0 inventory cars.

Much easier to convince a buyer to buy an AP 1.0 car with features the 2.0 cars still don't have.

I disagree. The reason AP2 cars don't have any features yet is because the fleet is still learning. You need cars with AP2 hardware out there collecting data to create the models to enable AP. Tesla couldn't do this before the announcement. I don't know the number, but I bet Tesla needs a few million miles of data. At some point they had to make the switch.

Also, if I recall correctly, Tesla wrote part of the AP1 software as well. Part of the disagreement between Tesla and Mobileeye was how far Tesla was pushing the 1.0 hardware.

 

[AWDtsla]

I disagree. The reason AP2 cars don't have any features yet is because the fleet is still learning. You need cars with AP2 hardware out there collecting data to create the models to enable AP. Tesla couldn't do this before the announcement. I don't know the number, but I bet Tesla needs a few million miles of data. At some point they had to make the switch.

Also, if I recall correctly, Tesla wrote part of the AP1 software as well. Part of the disagreement between Tesla and Mobileeye was how far Tesla was pushing the 1.0 hardware.

Uhh no. AP2 doesn't have features because the software isn't done/isn't out of validation yet. Fleet learning is only a tiny part of the algorithm.

 

[andrewket]

Uhh no. AP2 doesn't have features because the software isn't done/isn't out of validation yet. Fleet learning is only a tiny part of the algorithm.

It's both. It's circular, and no, machine learning isn't a tiny bit. It's significant.

 

[S4WRXTTCS]

No, not yet.

8.1 will likely be the last significant change, and then will see some slight tweaking. But, I wouldn't expect anything major improvement past 8.1.

 

[ChadFeldheimer]

I disagree. The reason AP2 cars don't have any features yet is because the fleet is still learning. You need cars with AP2 hardware out there collecting data to create the models to enable AP. Tesla couldn't do this before the announcement. I don't know the number, but I bet Tesla needs a few million miles of data. At some point they had to make the switch.

Also, if I recall correctly, Tesla wrote part of the AP1 software as well. Part of the disagreement between Tesla and Mobileeye was how far Tesla was pushing the 1.0 hardware.

Perhaps a more elegant way to handle this transition would have been to temporarily include the AP1 Mobileye chipset along with the new AP2 NVIDIA solution - until the AP2 fleet has replicated AP1 features. That way, AP2 consumers can enjoy AP1 features while waiting for AP2 software drops.

I'm confident that Tesla had considered doing this - and it would be interesting to find out why they chose not to do it this way.

 

[andrewket]

Perhaps a more elegant way to handle this transition would have been to temporarily include the AP1 Mobileye chipset along with the new AP2 NVIDIA solution - until the AP2 fleet has replicated AP1 features. That way, AP2 consumers can enjoy AP1 features while waiting for AP2 software drops.

I'm confident that Tesla had considered doing this - and it would be interesting to find out why they chose not to do it this way.

I had thought of this as well, but think of the cost of the extra hardware in tens of thousands of cars, not to mention the complexity of wiring the sensors to both chipsets at the same time. If Tesla really does get AP2 cars to parity of AP1 by 12/31 as they said they would, I think most owners will give Tesla a pass. My bet is that Tesla is late on this promise.

 

[182RG]

include the AP1 Mobileye chipset along with...

I'm confident that Tesla had considered doing this - and it would be interesting to find out why they chose not to do it this way.

The end of the relationship with Mobileye was more of a divorce, than a parting as friends. I think this was chosen for them at contract end.

The rush to substitute AP2 hardware, without software was another clue. I'm amused by the whole AP2 event. December is here. I especially love the spin around needing the new hardware to collect data for the "fleet learning" ahead of the software launch. Does anybody even know if the hardware is "actually" collecting data and forwarding to Tesla, or is it another EM smoke screen?

 

[croman]

Elon was very public in his statement. Only a fool wouldn't hedge and get called out like that. I think 8.1 will come in December and it will have some AP1 features. I don't think all will be included (but I'd expect TACC at minimum).

 

[u00mem9]

Elon was very public in his statement. Only a fool wouldn't hedge and get called out like that. I think 8.1 will come in December and it will have some AP1 features. I don't think all will be included (but I'd expect TACC at minimum).

Well I'm glad you said it so I wouldn't have to...

Re: fleet learning. You guys realize that same story was told about AP1.0 right? I mean fool me once...

 

[cab]

Agree that AP1 has likely seen the last of material enhancements. Bug and significant safety "fixes" will likely be it. While the OTA updates are nice and make Tesla cars "upgradeable", there is an obvious limit in terms of hardware and simply the manufacturer's resources measured against ROI. Heck, the browser is still mostly useless after 4 years, text message integration is non-existent, and Apple Carplay /Android Auto are nowhere to be seen (beyond one comment from Elon around how the latter would likely be better than the Tesla specic apps) - all odd from what many consider to be a "tech company".

 

[animorph]

I would expect no new features for AP1 unless there are some simple safety fixes possible. But I would think Tesla could continue training the AP1 system to perform better at what it does do. Plenty of AP1 cars driving around still.

 

[u00mem9]

I think we should play a game called: Guess the next functionality to be removed from Autopilot!

I'll start. No AP engagement on non-divided highways.

If we can list out the most likely candidates we can start a poll. Other obvious downgrade opportunities include:

No AP operation above the posted speed limit including divided highways

No automated lane change with turn signal (since ultrasonic sensors can't ensure safe clearance)

No autopilot use at low speed traffic jam conditions (only TACC)

 

[AWDtsla]

No automated lane change with turn signal (since ultrasonic sensors can't ensure safe clearance)

Probably this one. Incredibly dangerous to let the car merge for you, the only use of it now is to be able to change lanes without disabling AP. The novice/non-techy driver probably wouldn't understand the car is blind to the side and rear and should not be relied upon to merge.

We have until the first publicized accident until it is disabled.

 

[lunitiks]

A bunch of speculative whining!

In the 8.0 release notes there's a statement about upcoming "off ramp"-features in 8.1:

“When at a highway exit, it will favor marked lane lines in the direction of vehicle indicator (8.0)* or in the direction suggested by the navigation system if active (8.1). This feature will be available initially in the US and later in all countries where lane markings are open for highway exits.”

The way I see it, this may be a first step in pulling together AP and NAVI - which would be totally awesome!

Also - don't forget this blog post where Tesla explains how they're focusing more and more on the radar technology in combination with fleet learning.

I choose to think/wish we'll see more improvements in AP1.0 in the future, all though of course I realize that AP2.0 will advance much, much faster and become much, much better in driver assistance than AP1.0 - short and long term.

 

[chillaban]

A bunch of speculative whining!

In the 8.0 release notes there's a statement about upcoming "off ramp"-features in 8.1:

“When at a highway exit, it will favor marked lane lines in the direction of vehicle indicator (8.0)* or in the direction suggested by the navigation system if active (8.1). This feature will be available initially in the US and later in all countries where lane markings are open for highway exits.”

The way I see it, this may be a first step in pulling together AP and NAVI - which would be totally awesome!

Also - don't forget this blog post where Tesla explains how they're focusing more and more on the radar technology in combination with fleet learning.

I choose to think/wish we'll see more improvements in AP1.0 in the future, all though of course I realize that AP2.0 will advance much, much faster and become much, much better in driver assistance than AP1.0 - short and long term.

Please note that the original 8.0 blog post had a much less nuanced description of exit following. It used to say something more to the effect of using auto lane change to take highway exits. They eventually refined the wording to say that it will prefer lane markings towards an exit, which very rarely works for me.

 

[u00mem9]

A bunch of speculative whining!

In the 8.0 release notes there's a statement about upcoming "off ramp"-features in 8.1:

“When at a highway exit, it will favor marked lane lines in the direction of vehicle indicator (8.0)* or in the direction suggested by the navigation system if active (8.1). This feature will be available initially in the US and later in all countries where lane markings are open for highway exits.”

The way I see it, this may be a first step in pulling together AP and NAVI - which would be totally awesome!

Also - don't forget this blog post where Tesla explains how they're focusing more and more on the radar technology in combination with fleet learning.

I choose to think/wish we'll see more improvements in AP1.0 in the future, all though of course I realize that AP2.0 will advance much, much faster and become much, much better in driver assistance than AP1.0 - short and long term.

You posted the latest version of a claim that has been hyped no less than 3 different times, and your selection STILL claims things that aren't there. I get that the glass is 10% full in your mind, but to say there has been any connection between the blog posts and the actual capability gained is laughable