FSD may require a hardware upgrade...

[daktari]

You need auto wipers for all level 3 and above, aka FSD. Should be highest priority.

It is the sum of all the small hassles that make EAP not close to a finished product. Fast progress does not compensate for 100eds of bugs/lacking features.

 

[aWalkingShadow]

2) FSD video turns out to be a fraud.

2 seems like an assumption and a big one. think back to march 2016 when they've shown m3 prototype, would you call them a fraud in Mar 2017? never forget that there is no ETA for FSD and they dont owe us anything. i'm going to be really upset if there is no lvl3 by 2020 in m3 with FSD but even then, they're still not obligated since we bought without ETA being set.

I don't think we're making too big of an assumption. When you look at the disengagement data released by California, it shows lots of disengagements within a relatively few number of miles driven, which heavily implies that Tesla drove the same route multiple times until they got a good take. This, compounded with the fact that AP2 is nowhere near as capable as the demo from almost 1 year ago makes most of us think that something is off about the video. At best, the demo is misleading about their current state of technology, at worst it's fraud.

 

[dknisely]

My view:
- EM thinks FSD is possible and they have a team working on it. BTW, a smaller NN model is progress, it means they have made it more efficient -- same functionality, less memory. Now they can grow it again with new functionality.
- AP1 = Mobileye object-recognition + Tesla driving-logic.
- AP2 = Tesla NN object recognition + Tesla driving-logic.
- FSD = deep NN + integrated mapping and route-planning.
- AP1/AP2 and FSD are separate teams. Features might cross-over, but would require separate development, or the two codebases running in parallel.
- Automatic wipers are nice, but I have survived my entire life without them -- they are a low-priority goal for teams tahh have much else on their plates.
...

All the focus is on getting the 3-ramp done. Most hw and sw teams will be focused on that. I thin kit is going better than I had expected.

I guess you never drove AP1 with auto-wipers. You have missed out, esp. from the location you seem to be from.

 

[dknisely]

They'll get get fsd done for current owners of ap2, ap2.5. the quiet period is just a detour. Don't forget company other big projects which may have taken focus from fsd/Tesla network. Model 3 (you mentioned), 100 day battery install in Australia, and restore fsd team morale / fix turnover. I would not be surprised if engineers left because of leadership that was promptly removed (suppose not promptly enough). Having made their excuses, ì'd like my m3 to drive itself from factory to my garage on fsd today.

Okilly, dokilly. I'll settle for auto-wipers and AP2 that doesn't lunge into things on local roads.

 

[u00mem9]

Side note: demo video was also sped up which obscures the fact that the car stops in 3 odd places any one of which could easily have resulted in being rear ended (especially if the following car was a newer AP2 tesla instead of an Mobileye equipped car)

 

[alcibiades]

My view:
- EM thinks FSD is possible and they have a team working on it. BTW, a smaller NN model is progress, it means they have made it more efficient -- same functionality, less memory. Now they can grow it again with new functionality.
- AP1 = Mobileye object-recognition + Tesla driving-logic.
- AP2 = Tesla NN object recognition + Tesla driving-logic.
- FSD = deep NN + integrated mapping and route-planning.
- AP1/AP2 and FSD are separate teams. Features might cross-over, but would require separate development, or the two codebases running in parallel.
- Automatic wipers are nice, but I have survived my entire life without them -- they are a low-priority goal for teams tahh have much else on their plates.
- Choice was: add AP2 hw to all cars, or face having a large number of cars not AP2/FSD capable. They chose the best hw available for an affordable price. hw improves all the time, and installed hw will change. The processing unit is easily retrofittable. *If* hw2.5 is needed, it can be retrofitted into the (relative to number of Model 3s) small number of S/X.
- Development trials during October were mainly debugging the hacked on mapping and route-planning software onto the NN model. The November demo had no interventions. I say that is a quite effective *demo* -- but always take demos with a grain of salt -- they usually reflect a specific set of circumstances. The difficulty with true FSD is generalizing it to most situations.

I think that we should put ourselves into the shoes of Tesla. They are juggling sw and hw development. They took a risk by installing AP2 hw into *every* car. Sw development is always fraught with delays -- this is especially true for NN development, which is on the bleeding-edge. I feel that EM is being honest in his foresight given what he kwows/knew at the time of his predictions. He is overly optimistic in his time estimates, and we can account for that, but I think his optimism has actually sped development.

All the focus is on getting the 3-ramp done. Most hw and sw teams will be focused on that. I thin kit is going better than I had expected.

I can't recall. Which HW do you drive with presently?

 

[Swift]

(especially if the following car was a newer AP2 tesla instead of an Mobileye equipped car)

Nice.

 

[buttershrimp]

Don't take AP2 out behind a barn to hit it with a shovel... give them till the end of the year to start integrating things. There is an impressive team at Tesla that is working on it. As for the AP1 mantra, IMHO, AP2 is pretty much at parity now.

Would love someone to post me a video doing the exact same course of road with both versions of a current loaner, back to back.... if someone can be objective and not try to prove their existing viewpoint.... AP2 will be doing some surprising things by the end of the year.

They will make the deadline to demonstrate the cross country drive....

As for all the AP2 haters.... and myself (an AP2 optimist beginning to yearn for more aggressive updates)... ... whoever is wrong should come back a year from now and say one of two things:

"FSD was a pipe dream. AP2 blows and AP1 still kicks ass" or on the other side... "Suck it y'all.... my AP2 car is exiting and entering highways, and parking in tight spaces and auto sensing the hell out of some rain droplets!"

Let's give it time for the nerds that are working hard to get it done.

 

[AnxietyRanger]

Don't take AP2 out behind a barn to hit it with a shovel... give them till the end of the year to start integrating things. There is an impressive team at Tesla that is working on it. As for the AP1 mantra, IMHO, AP2 is pretty much at parity now.

I just love your humor posts, @buttershrimp! Keep up the good work.

 

[oktane]

No, I thought EAP was expected December 2016, not just "parity". Tesla is keeping quiet since they know there is nothing they can say to salvage this. EAP will never reach parity with AP1.

That being said, I'd favor a closed course AP1 vs. AP2 video. It may be hard to judge subjective thinks like jerkiness.

 

[chillaban]

That being said, I'd favor a closed course AP1 vs. AP2 video. It may be hard to judge subjective thinks like jerkiness.

That’s a huge problem. Some things like a sudden 5mph deceleration look fine on camera but feel rather jolting in the cabin and will get your adrenaline pumping. Other thing like slight side to side wobble also don’t translate on video.

I had a driving instructor who told me that he would rather me straddle the lane line for seconds rather than suddenly snap the car back to center. I guess he was running AP1.

I would love to see targeted tests of AP1 vs AP2 across many runs on a similar course, preferably at the same time. But even that might not answer all of our questions.


FWIW after coming back from another 800 mile trip, 2017.28 AP2 did pretty well compared to my early 17.x.x experiences on AP1. Slightly more overpass braking events and more concerning a few instances of truck butt lust (can I officially coin that term? You’re welcome....) where it wanted to accelerate towards a truck that just entered my lane. But OTOH it managed to navigate more of the sweeping curves between Barstow and Bakersfield that AP1 was never able to. AP1 would happily and “confidently” depart my lane right towards neighboring cars during those curves, but even 2017.28 successfully tracked them.

Overall, I would say at this point AP2 is showing a faster reaction time to rapidly changing conditions, but with that comes a bunch of jumpiness from sudden changes in its assessment of the world, most of which unnecessary. For a stretch of CA-99, turning lanes widened the left lane just slightly. AP2 would snap towards the “new” center and then snap back to the old center every time, when a human driver would’ve just ignored the momentary changes in lane width.

Overall though, AP2 does seem to be rapidly improving in capability. It’s still got a ways to go, but the progress is quite noticeable in pace.

 

[u00mem9]

... whoever is wrong should come back a year from now and say one of two things:

That's what the EVJ disciples said about a year ago...how many years should people wait to conclude it's a giant cluster?

Personally, I plan to wait until someone is repairing an AP2 car and reveals that 1/2 the cameras aren't even connected to anything.

 

[BigD0g]

That's what the EVJ disciples said about a year ago...how many years should people wait to conclude it's a giant cluster?

Personally, I plan to wait until someone is repairing an AP2 car and reveals that 1/2 the cameras aren't even connected to anything.

Umm, we know that's not true @verygreen has already posted numrous captures of the 8 cameras data collection. However, I don't disagree that it's a giant cluster, but it most certainly is.

 

[lunitiks]

They're connected. I've seen it up close. Have a peek at @kdday's excellent "Inside" the PX2-thread, where we all got our first (and so far only) good look at the ECU assy.

Man that was an exciting day

 

[BigD0g]

They will make the deadline to demonstrate the cross country drive....

Care to make a bet on this one? Assuming the deadline is what Elon stated as Dec 2017, and we are not adjusting for "Elon Time" ?

 

[u00mem9]

Care to make a bet on this one? Assuming the deadline is what Elon stated as Dec 2017, and we are not adjusting for "Elon Time" ?

This forum cannot agree that the last video was a fake, betting on the next one is likely to descend into the same nonsense.

 

[BigD0g]

This forum cannot agree that the last video was a fake, betting on the next one is likely to descend into the same nonsense.

I'd go even further saying, there won't even be a video in Dec 2017, so if there is an actual coast to coast video posted in Dec 2017, you win, even if it's completely fabricated. Frankly, I love my odds .

Given it's now Sept. we just received the .34 release with no new feature updates, it still can't take curves without drifting to outside lanes. Realistically given Tesla's trajectory for wide releases we have probably 3 - 4 releases left before 2017 at best.

 

[d21mike]

I ordered my car last week and I may get it by the end of this month so no experience. I have watched a lot of videos. I think not stopping at a Red Light while on Auto Pilot is the main change I would like to see. And for a small change I would like it to turn off the blinker automatically after changing lanes. I am ok with incremental changes more quickly then waiting for a lot of fixes all at once which seems to be what they are doing.

 

[BigD0g]

I ordered my car last week and I may get it by the end of this month so no experience. I have watched a lot of videos. I think not stopping at a Red Light while on Auto Pilot is the main change I would like to see. And for a small change I would like it to turn off the blinker automatically after changing lanes. I am ok with incremental changes more quickly then waiting for a lot of fixes all at once which seems to be what they are doing.

Oh, welcome to the club! However, there is A LOT of things we need to see before working about stopping at stop lights and turning the blinker off. You'll see when you have your car, we need better driving before worrying about creature comforts and stopping at stop lights I "highly" doubt will come before highway exit ramps or transitioning highways, as one is an EAP feature and the other is an FSD feature.

 

[d21mike]

It still can't take curves without drifting to outside lanes.

I have seen a lot of Tesla Owner Youtube Video's that disproves this. Of course since I have not received my car yet I have zero personal experience. Maybe if you had said "can't take all curves at any speed" might be more accurate.