Tesla autopilot HW3

[electronblue]

Why so? Even at a $2k added, swapping 6 camera modules at the same time they swap the AP module does not seem difficult or a negative cost.
Swapping the radar would a bit more of a pain due to the bumper cover, but trained mobile service teams could handle it quickly and efficiently.

Thank goodness for diversity of people and opinion, I guess.

Here we are many of us feeling pessimistic about even receiving the HW3 computer upgrade at all after the firesale of FSD upgrades these last few days...

...and along come others who feel even a sensor upgrade might be coming!

At least there is something nice about that thought.

 

[mongo]

Cost of time (+opportunity) , labor, parts will exceed the Musk'd $2k FSD price.

There are companies that only do this kind of thing (rework) no opportunity cost. 1k for AP computer + $600 for cameras plus two hours of labor ($300) and drive time ($100) = $2k.

It's less work than an engine swap...

and along come others who feel even a sensor upgrade might be coming!

To be clear, I'm not saying it will happen, but that if it did need to, it would not kill Tesla or be overly burdensome.

 

[iTech]

There are companies that only do this kind of thing (rework) no opportunity cost. 1k for AP computer + $600 for cameras plus two hours of labor ($300) and drive time ($100) = $2k.

It's less work than an engine swap...



To be clear, I'm not saying it will happen, but that if it did need to, it would not kill Tesla or be overly burdensome.

What kind of company works for free?

 

[mongo]

What kind of company works for free?

No opportunity cost since Tesla would not need to devote resources to it that could be used elsewhere. The contracted company charges, of course, but the prepayment on FSD/EAP is more than the replacement cost. Can even view it as good will and take a minor hit on the margin of EAP/FSD of future sales.

 

[J1mbo]

Why so? Even at a $2k added, swapping 6 camera modules at the same time they swap the AP module does not seem difficult or a negative cost.
Swapping the radar would a bit more of a pain due to the bumper cover, but trained mobile service teams could handle it quickly and efficiently.

The AP2.5 radar has 2 feeds, whereas AP2.0 has 1 feed. That could be a big deal for a retrofit (assuming HW3 has the 2.5 radar and not something else).

 

[mongo]

The AP2.5 radar has 2 feeds, whereas AP2.0 has 1 feed. That could be a big deal for a retrofit (assuming HW3 has the 2.5 radar and not something else).

You're talking about the dual CAN bus connections? Yeah, some of the hardware (steering rack) has gotten more fault tolerant over time.

 

[iTech]

No opportunity cost since Tesla would not need to devote resources to it that could be used elsewhere. The contracted company charges, of course, but the prepayment on FSD/EAP is more than the replacement cost. Can even view it as good will and take a minor hit on the margin of EAP/FSD of future sales.

You're missing the opportunity cost of having to train that external company and then making sure they don't actually cause damage.

The AP2.5 radar has 2 feeds, whereas AP2.0 has 1 feed. That could be a big deal for a retrofit (assuming HW3 has the 2.5 radar and not something else).

Also eFuses

 

[mongo]

You're missing the opportunity cost of having to train that external company and then making sure they don't actually cause damage.

That is minimal, and the contract would likely have a clause for damage.

Also eFuses

having an eFuse doesn't help while there is a short though.

 

[dgatwood]

This seems like they are moving in that direction, or at least wanting to keep that door open.

Or rather, cut a hole in that brick wall so that they can install a door there later, if they decide they need to... now I'm really stretching the analogy.

Basically they're starting to realize they've built up a rather large legal/financial liability with AP2 and they are working on ways to cut their losses, ideally to zero, just like they did with AP1, and move on to AP3 (which is not to be confused with HW3; I suspect AP3 will include a much better sensor suite with more redundancy.)

Ignoring missing redundancy, there's no reason you couldn't trivially retrofit a better sensor suite into the existing cars. It's not like the AP1 to AP2/2.5 change where they went from one camera to 8, which would have required completely dismantling the car from one end to the other just to add the wiring harness.

Adding LIDAR to existing vehicles, if that proves necessary, would be a simple matter of ripping out the RADAR unit, swapping out the front bumper/bumper cover with a modified version that accommodates the LIDAR hardware, and building a custom LIDAR package that connects to the same cable that the RADAR did. And ostensibly, corner RADAR units could be added in place of the existing SONAR with only minor local modifications as well.

 

[electronblue]

Ignoring missing redundancy, there's no reason you couldn't trivially retrofit a better sensor suite into the existing cars. It's not like the AP1 to AP2/2.5 change where they went from one camera to 8, which would have required completely dismantling the car from one end to the other just to add the wiring harness.

Adding LIDAR to existing vehicles, if that proves necessary, would be a simple matter of ripping out the RADAR unit, swapping out the front bumper/bumper cover with a modified version that accommodates the LIDAR hardware, and building a custom LIDAR package that connects to the same cable that the RADAR did. And ostensibly, corner RADAR units could be added in place of the existing SONAR with only minor local modifications as well.

I am not sure what is the usefulness of these mental exercises as there is no history of Tesla ever retrofitting past Autopilot suite upgrades to old cars. Those LIDAR changes alone would require designing a way for the LIDAR to see forwards ie bumper changes or at least changes to the ”grille” parts...

Even the one exception that Tesla has discussed, HW3 computer to HW2/2.5 cars, seems iffy now that the amount of potential retrofits has grown so much due to the recent firesale.

In the past Tesla has simply moved on from broken Autopilot promises and the only recourse has been to buy a new car.

 

[dgatwood]

I am not sure what is the usefulness of these mental exercises as there is no history of Tesla ever retrofitting past Autopilot suite upgrades to old cars. Those LIDAR changes alone would require designing a way for the LIDAR to see forwards ie bumper changes or at least changes to the ”grille” parts...

Tesla, back when AP1/HW1 was still for sale, also did not promise that if new hardware was required, they would upgrade it for free. They did make that promise beginning with AP2/HW2. Based on that promise, people bought cars that they otherwise would not have purchased without that promise. That makes it a very, very different legal liability story for Tesla if they decided to back out. There's no way Tesla could back out without it costing at least 10x what the upgrade would cost them, and quite possibly 100x or more. Not to mention that any customers so burned would never trust Tesla again.

Suggesting that Tesla would screw over nearly all of their owners like that is, frankly, absurd. If they did that, their lawyers would probably be disbarred for gross incompetence for letting them do so, and it would almost certainly mean the end of the company in a single lawsuit.

Not gonna happen.

Even the one exception that Tesla has discussed, HW3 computer to HW2/2.5 cars, seems iffy now that the amount of potential retrofits has grown so much due to the recent firesale.

In the past Tesla has simply moved on from broken Autopilot promises and the only recourse has been to buy a new car.

It's probably a $500 dollar board swap. No way are they going to risk a major lawsuit and a settlement so big it would bankrupt the company just to save a few hundred bucks per car.

Even if we go with the most absurd estimate and assume that everybody bought full self-driving, that would still only be about 370,140 cars plus whatever they've sold this year, times a few hundred bucks each, or O($185M). And in reality, only about 40% of people prepaid for FSD even when it was available at purchase time. I'd imagine the numbers during the period where it had to be manually added were close to zero. So I doubt they'll have to do more than about 120k upgrades, at a cost of $60M dollars plus labor, which is less than Tesla's burn rate for an average week in 2018.

The cost of the upgrades is noise. Even when you add labor costs in, it's still noise.

 

[electronblue]

Tesla, back when AP1/HW1 was still for sale, also did not promise that if new hardware was required, they would upgrade it for free. They did make that promise beginning with AP2/HW2. Based on that promise, people bought cars that they otherwise would not have purchased without that promise. That makes it a very, very different legal liability story for Tesla if they decided to back out. There's no way Tesla could back out without it costing at least 10x what the upgrade would cost them, and quite possibly 100x or more. Not to mention that any customers so burned would never trust Tesla again.

Of course not but Tesla did promise many software features that never appeared for AP1, such as the announced navigation-based exit-taking for firmware 8.1 in December 2016 and many other feature promises they made over that period of time — all now forgotten.

Here’s the thing. Tesla has already started to rephrase what FSD means in their public communications. When many of us bought FSD in 2016 and 2017 there was the fresh talk of ”Level 5 capable hardware” and ”Tesla Network details in 2017” for autonomous ride sharing but now Elon is on the record calling NoA as FSD on the highway...

Tesla will surely deliver something called FSD for AP2/2.5 cars but it seems far from certain that outcome is the one people originally had reason to expect. Given this background it seems like a plausible scenario that Tesla would do the bare minimum (retrofit-wise) to reach this and then move on when new hardware platforms and suites are introduced.

Tesla never promised hardware upgrades to AP2/2.5. They said a computer upgrade would be made if necessary. All they have to do is say it is not necessary...

Suggesting that Tesla would screw over nearly all of their owners like that is, frankly, absurd. If they did that, their lawyers would probably be disbarred for gross incompetence for letting them do so, and it would almost certainly mean the end of the company in a single lawsuit.

Not gonna happen.

Tesla got away with the P85D HP debacle, all those AP1 features undelivered, P90D Ludicrous missing its performance targets many times over and so forth. There were some minor, localized lawsuits but nothing major. Tesla promised lighted vanity mirror retrofits to all Model S cars — never happened. And the train moved on to new products solving the issues for new customers... Why would AP2/2.5 necessarily be any different from this past?

I’m still holding out hope that HW3 or somesuch computer retrofits will happen for AP2/2.5 but it seems far from certain. As for any sensor suite upgrades in my view those sound completely unrealistic.

 

[verygreen]

They would still end up writing code for AP2 cars to convert the colorspace to match AP2.5 right

You cannot fully correct the colorspace. you would not be able to tell blue from green apart. Probably not as important in US, quite important in other parts of the world.

 

[acoste]

According to Electrek:

"We contacted Tesla to clarify the situation about Hardware 3.0 and the company says that the new Autopilot computer is still not shipping in new cars."

https://electrek.co/2019/03/06/tesla-self-driving-language-walks-back-features-confusion/

 

[iTech]

According to Electrek:

"We contacted Tesla to clarify the situation about Hardware 3.0 and the company says that the new Autopilot computer is still not shipping in new cars."

Tesla updates self-driving language, creates confusion and walks back features

What happens to that new NN then?

 

[electronblue]

What happens to that new NN then?

we are f-cked....

?

 

[iTech]

?

I'm confused on what you're asking?

 

[electronblue]

I'm confused on what you're asking?

I was asking whether @BigD0g ’s quote might be the answer to your question?

 

[iTech]

I was asking whether @BigD0g ’s quote might be the answer to your question?

Haha, yeah I would agree with that response as the answer.

 

[mongo]

If the FSD option price is more than the cost to swap HW2.5 for HW3, then it is a no brainer to use up the sunm cost of current inventory and supply agreements for HW2.5.
Best case, some people don't get FSD and HW2.5 stays in the car. Worst case, everyone get FSD, and there are a lot of swaps, but that happens in a future quarter. Also gives time to vet HW3 against NN(whatever number it is).

Question: what would swapped HW2.x modules be good for?