Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

HW2.5 capabilities

This site may earn commission on affiliate links.
1. Redundancy shouldn't affect day to day performance of the car. It's only critical when something fails.
I would think that redundancy would be useful in day to day if weather or lighting (day/night/glare/sunset/headlight limits) was included. As an example in rain or fog radar may be able to detect pedestrians (etc) when cameras could not (at least at their normal precision). Similarly with the different lighting conditions.

I would think in general that multiple types of sensors would add to the confidence level of identifying all types of objects. Just a thought.

Aside: in my post above I meant '...quite a distance...' and not '...quite a difference...'
 
My thoughts -

1. Redundancy shouldn't affect day to day performance of the car. It's only critical when something fails. This means that AP2.0 user can get the same performance of FSD, but they have a higher chance of failure rate.

2. Hoping its actually August 21st. Oct 21st 2017 is a Saturday so don't know if that actually makes sense. (Oct 16th 2016 was a Monday as well).

When they promised FSD in AP2 would be twice as safe as the average driver, Tesla will simply claim in the response to the lawsuit that they used a sample of octogenarians who were simultaneously attempting to facetime with their grandchildren
 
  • Funny
Reactions: Matias
I still think that there's a lot of hand waving going on with fsd and eap for that matter. Redundancy is another matter (it's easy enough to count One Camera), but as far as computing power needed for eap and fsd it reall is an unknown. As in it isn't known because no one has done it before. And frankly, I have no confidence in the position that Tesla has some magic FSD solution running somewhere that could clue them into the scale of processing power needed. There's just no evidence of anything sophisticated enough to warrant a belief of NN magic happening. We haven't seen it in EAP nor have we seen any new videos.

I think it's possible they're doing basic dsp edge finding and curve fitting for lane lines and then some algorithmic process for staying in the lane. I think they will eventually start using NN processing for control, but it's not now.

The history of computer science is full of radically wrong estimations of capabilities and capacities - both to the good and bad. The AI of 2001 still isn't around (a relief for those of us in deep space hibernation) and on the other hand a computer mastered Go much sooner than anticipated. Nevertheless, the data from you guys is fascinating!
 
What? The date in the parts list?
958a278534504f30e1d04fce2c50818535f926c9e6f0aa5f6fa800f0e3799ad3.jpg
 
So clearly, whatever the changes Tesla referred to as "2.1" in the Electrek statement began building August 21st - the redundant wiring etc.

Interesting to see if this matches what we think is in the Model 3 or if this is some intermediary step on the road to a still larger overhaul (e.g. does the interior camera wait for an interior redesign etc.).
 
I also find the e-fuse part interesting.
Me too. The new e-fuse doesn't replace any of the fuse boxes in the car, but adds to them.

According to the schematics, e-fuse protects 10 critical wires: Power for (from?) the DC-DC converter, power for the frunk fuse box, the cabin fuse box, primary power steering, redundant power steering, air suspension compressor, condenser fan right, condenser fan left, ABS motor and iBooster (power braking).

In short: Probably all critical driving systems.

So why. WhhhyyyyyYYy this e-fuse. Isn't Model S already the safest car in the world?

Well, there's also logic interface on this thing :). The logic interface controls the e-fuse box itself ("ENABLE E-fuse" etc.) and seems to communicate with both the DC-DC-converter and the BMS ("ENABLE HV Batt").

So I guess a more appropriate name for the e-fuse would be Remote Kill Switch :)
 
Me too. The new e-fuse doesn't replace any of the fuse boxes in the car, but adds to them.

According to the schematics, e-fuse protects 10 critical wires: Power for (from?) the DC-DC converter, power for the frunk fuse box, the cabin fuse box, primary power steering, redundant power steering, air suspension compressor, condenser fan right, condenser fan left, ABS motor and iBooster (power braking).

In short: Probably all critical driving systems.

So why. WhhhyyyyyYYy this e-fuse. Isn't Model S already the safest car in the world?

Well, there's also logic interface on this thing :). The logic interface controls the e-fuse box itself ("ENABLE E-fuse" etc.) and seems to communicate with both the DC-DC-converter and the BMS ("ENABLE HV Batt").

So I guess a more appropriate name for the e-fuse would be Remote Kill Switch :)

Wouldn't have thought they needed to go to this extent for a remote kill switch, unless they are worried about a rogue AP AI trying to take over the world. In which case an eFuse connected to the CAN probably won't help much ;)

Guessing they have found a scenario where the car doesn't (cannot?) completely shut down after an accident, or maybe this is a step towards ditching the 12V battery?
 
I don't think they will kill the 12v battery until the S/X Refresh in 2020 or so, but clearly AP2 is now on very very shaky ground. The fact that they are switching radars to a better one with more find grain course plots and better able to detect pedestrians, a more redundant wiring harness and of course the dual gpu parker box. We may not be at a spine transplant, but we are certainly getting close to a nose job and a tummy tuck.
 
I still think that there's a lot of hand waving going on with fsd and eap for that matter. Redundancy is another matter (it's easy enough to count One Camera), but as far as computing power needed for eap and fsd it reall is an unknown. As in it isn't known because no one has done it before. And frankly, I have no confidence in the position that Tesla has some magic FSD solution running somewhere that could clue them into the scale of processing power needed. There's just no evidence of anything sophisticated enough to warrant a belief of NN magic happening. We haven't seen it in EAP nor have we seen any new videos.

I think it's possible they're doing basic dsp edge finding and curve fitting for lane lines and then some algorithmic process for staying in the lane. I think they will eventually start using NN processing for control, but it's not now.

The history of computer science is full of radically wrong estimations of capabilities and capacities - both to the good and bad. The AI of 2001 still isn't around (a relief for those of us in deep space hibernation) and on the other hand a computer mastered Go much sooner than anticipated. Nevertheless, the data from you guys is fascinating!

I'm quite certain that they're using NNs to determine the relative position of the car in the lane. Given the number of highly successful examples of people doing that, I can't imagine it being otherwise.

But I am equally concerned that they're perhaps getting ahead of themselves. I don't believe FSD is going to be nearly as easy as people think. I still say we're looking at 20 years for widespread availability of true FSD. And it will require infrastructure upgrades.

I expect that we're going to see a slow introduction of FSD lanes, FSD highways/roads, as they slowly standardize what's required to safely implement FSD. And I think we're a long way from allowing FSD in mixed urban road/ pedestrian traffic.

All of which does not mean that L2,L3,L4 aren't going to be useful and enhance safety, well ahead of true L5 FSD. It's all good stuff. But we need to be realistic.
 
I don't believe FSD is going to be nearly as easy as people think. I still say we're looking at 20 years for widespread availability of true FSD. And it will require infrastructure upgrades.

Agree, as does Amazon who holds (or at least filed) some infrastructure related patents last year - presumably to remove drivers from their final delivery leg - in case the drones don't work out for large items ;)

I think less than 20 years tough. As Gates is credited with, "most people overestimate the change that will occur in the next two years and underestimate the change that will occur in the next ten".

We'll see.
 
  • Like
Reactions: Rouget and Rossy
Ladies and gentlenerds, the cat has escaped the bag: We have new wiring schematics for Model S! Woo-hoo!

Interestingly they're called "October 21, 2017"... Which is two months from now... So naturally there's some speculation that "October" could be a typographical error and that it should've read "August".
Since August 21, 2017 was yesterday.
And since the previous wiring diagram was released in "October" 2016.
And since Tesla has never released anything that early before.

Who knows. Fatty-fat fat fingers, maybe.

Anyway: Autopilot "HW 2.5" - yes: Tesla actually calls it that in the schematics - seems to include the following changes:

1) "Gen3" Power Steering ECU with redundant power supply from a new E-fuse box in the frunk. (The E-fuse box itself is a pretty big HW-change as well, but I won't go into that.)

2) Rear view camera wiring/connectors has changed in the Autopilot ECU. Video signal still goes directly from rv-camera to the ECU (same as ap2.0), but now there's a "splitter" on the ECU side that for some reason sends signal out to both the MCU and to the ECU itself! So now there's two BUC inputs; one "Splitter Rearview Cam IN" and one "APE Rearview Cam IN". Don't ask me why. Anyway cudos @verygreen for noticing this in the SW-code long ago.

3) New radar sensor from Continental ("Radar Front Conti") with "primary" and "secondary" CAN-signalling to AP2.5ECU. Exciting!

4) Different fuse location for the 30A Autopilot ECU Primary power (F133 > F214). The 10A Autopilot ECU Secondary power is same as before. Not so exciting...

Non AP-wise, I also noticed a new "Charge Port ECU" that was not in the 2016-diagrams at all. (WTH?)

Well that's about it. No interior/cabin camera and still no HUD for you.

On a slightly different note, do we have any indication whether the 75 cars now get 85 BTX8 batteries in the US?
 
Wouldn't have thought they needed to go to this extent for a remote kill switch, unless they are worried about a rogue AP AI trying to take over the world. In which case an eFuse connected to the CAN probably won't help much ;)
All of these changes are consistent with L5 and ride-sharing and align with what Tesla told Electrek.

A remote kill switch is for a ride-sharing passenger to intervene in the event of an emergency.

Wiring and other redundancies are to meet the "redundancy" requirement of L5, which is a nebulous requirement that's not yet fully fleshed-out.

The only question in my mind is whether they're also putting an internal camera in the S/X to monitor the driver and cabin.
 
  • Helpful
Reactions: AnxietyRanger
Well that would be interesting, wouldn't it. Especially since the FSD-disclaimer has always said that it is dependent on SOFTWARE validation. Software. Not hardware upgrades. I'm going to quote @AnxietyRanger again: We shall see.
Woulndt be the first time tesla doesnt deliver in AP.
For AP1 tesla said the car would navigate in your private driveway to park itself.
We still dont have that.
 
Similarly ape (the HW2.x computer) code has a bunch of checks in the form of "am I on hw2.5? if so, am I on model 3? don't ask mcu to init the backup camera. Not model 3? Call this other script to init backup camera".
So why would the M3, with AP2.5, not "intialize the rearview camera?" Or does this imply that initialization for that model is not required.

One possible interpretation if the above quote is correct is that Tesla won't be using the rear view camera on the M3 for AP functions. I hope this isn't the case!
 
  • Love
Reactions: Cobbler