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

Firmware 9 in August will start rolling out full self-driving features!!!

This site may earn commission on affiliate links.
@Bladerskb
Please point us directly to the post where wk057 said that:
  • Vehicle detection to the sides and behind your vehicle is complete garbage....
It would surely be interesting to read.
I couldn't find the original post googling some parts of it.
Couldn't pick it up from a quick glance over his recent posts either.

See the initial post of this thread: Navigate on Autopilot is Useless (2018.42.3)

In general, it is a very interesting and critical review of the initial version of Navigate on Autopilot. Sadly many of the points still hold, despite what many Tesla defenders may claim.
 
An interesting read indeed.
I'm not surprised about many of these issues, most of these seem to be explained by the current absence of a proper driving strategy.
Some sanity checking seems to be remarkably absent, I wonder if they re-started from scratch when developing NoA.

However, concerning the computer vision I want to reiterate my claim that Teslas 18.48.12.1 as visualised in
seems about on par/if not a little better than MobileEyes currently in vehicles. Rest assured all vision suppliers have issues with ghost objects, distance, speed, yaw estimation. It just a matter of suppressing exactly the right amount of that before passing it along to the OEM :)

I'm more worried about the absence of traffic signs and traffic lights in these visualisations.
@verygreen didn't you find any to visualise? is there still data structures left you couldn't make sense of?

NoA in Europe seems to be held up due to "lots of unique road markings & signs in each country". No *sugar*!
And EAP seems to brake for mapped speed limits only as well (so do most other speed limit assistants as well even if they have traffic sign recognition on board).
 
  • Disagree
  • Informative
Reactions: Bladerskb and ChrML
I'm more worried about the absence of traffic signs and traffic lights in these visualisations.
@verygreen didn't you find any to visualise? is there still data structures left you couldn't make sense of?
Given that the HW2 computing power is just enough to handle AP in its current incarnation with 8 cameras, perhaps it's a conscious omission to minimise load, especially since it's not strictly even needed for a level 2 driver assist feature.
 
Given that the HW2 computing power is just enough to handle AP in its current incarnation with 8 cameras, perhaps it's a conscious omission to minimise load, especially since it's not strictly even needed for a level 2 driver assist feature.

If you don't believe this (ESPECIALLY for AP2.0) the way the AP handles the TACC portion of EAP vs just using TACC is like a driver texting and making a grocery list in stop and go traffic VS a driver paying full attention and being fully aware up to as far as the eye can see. I have gotten to the point of only using EAP for my daily commute when I am by myself and using ONLY TACC when my wife and/or kids are with me. The sudden brake pedal snap and abrupt braking as if not paying attention to what is going on just got too much to deal with when others are in the car. Hoping HW3 will fix this dramatically as my wife's HW2.5 is 90% better at doing both under EAP.
 
@boonedocks

just to clarify - you are saying that your AP2.5 car behaves significantly better on EAP than your AP2.0 car?
Haven't seen much evidence of this comparison. Interesting. Thanks.
100% yes where attentive braking is concerned. Lane following and NOA are equivalent but braking while using EAP is very different between my 2.0 & 2.5 cars.
 
100% yes where attentive braking is concerned. Lane following and NOA are equivalent but braking while using EAP is very different between my 2.0 & 2.5 cars.
That's fascinating, and very concerning. Unless people bought FSD they won't be getting the HW3 upgrade so people on HW2.0 will have a lesser setup. The cameras are also different between them with grey/red and full colour cameras in 2.0 and 2.5 so perhaps it's not even software and the camera differences... Have you considered that your cameras just may need checking?
 
  • Like
Reactions: MorrisonHiker
That's fascinating, and very concerning. Unless people bought FSD they won't be getting the HW3 upgrade so people on HW2.0 will have a lesser setup. The cameras are also different between them with grey/red and full colour cameras in 2.0 and 2.5 so perhaps it's not even software and the camera differences... Have you considered that your cameras just may need checking?
My service center has checked everything from top to bottom and even replaced the brake booster thinking it might be it. I have just decided to wait for HW3 and see if it’s magically fixed

35080026-91B9-40FC-AD2A-DD0FA2088FE0.png
 
However, concerning the computer vision I want to reiterate my claim that Teslas 18.48.12.1 as visualised in seems about on par/if not a little better than MobileEyes currently in vehicles. Rest assured all vision suppliers have issues with ghost objects, distance, speed, yaw estimation. It just a matter of suppressing exactly the right amount of that before passing it along to the OEM :)

This is simply not true as already been documented with comparison of raw inputs coming from the 2014 EyeQ3 (AP1) vs AP2 (November NN). But don't let me stop you from believing fairy tales and ignoring blatant facts.

giphy.gif
 
  • Disagree
Reactions: HolyGrail
Ok here is the full context of the quote that @Bladerskb posted from the earnings call today. There is actually a lot of information in Elon's answer. I painstakingly transcribed the entire audio manually:

Question: Can you please share an update on FSD and the Tesla Network development? When will customers start to see FSD features? What's the best case timeline for the Tesla Network to go live?

Elon: Sure. We already have FSD capability on highways so from highway on ramp to highway exiting, including passing cars, and from one highway interchange to another, FSD capability is there. In a few weeks we will be pushing an update that will allow the option of removing stalk confirm in markets where regulators approve it which I believe will be the case in the US for example and over time, I believe all regulators will approve it. But we kept stalk confirm just to make sure we took care of any strange corner cases. It's really quite sublime, you have stalk confirm off and, like, the car goes from highway on ramp to passing slower cars, takes an interchange then takes an exit and comes to a stop after the exit. So, it's actually quite profound to have that experience.

The next part of FSD will really be.. is traffic lights. It's hard. [garbled] streets is pretty easy. You can essentially geocurve[?] those and easily recognize stop signs. Traffic lights and intersections will be the next really tricky one and then navigating complex parking lots. Like if you are underground in a mall parking with a lot of traffic and pedestrians and it's on multiple levels. That sort of thing is where it gets tricky. With the release of enhanced or advanced summon, you will see the first indications of the car being able to navigate complex parking lots. And that is also coming out fairly soon, probably next month. And in development mode, the car does all the things I just mentioned, recognizes traffic lights, stop signs and all the functionality in development mode. It's really just about getting the reliability of recognizing traffic lights to several 9's. It's like maybe 98% good right now but we need to be 99.999%, really extremely reliable. When do we think it will safe for FSD? Probably by the end of this year and then it will be up to regulators to decide when they want to approve us.

Ok here is my answer to your question @Bladerskb . It appears that Elon considers NOA, either now or the version that will go out with no stalk confirm, to be FSD on highways because it can do the functions required to navigate a highway without driver input (ie take a highway ramp, pass a car, take an exit). I do not agree that NOA now is FSD. So no, I cannot defend that statement from Elon.

However, if NOA in a future update with no stalk confirm can do what Elon describes (ie drive on the highway, pass cars, take exits etc) reliably without any driver intervention or nags, then I would consider that to be highway self-driving.

A few extra observations. Based on Elon's answer, it appears advanced summon should come out soon (he says probably next month but it might be a few months from now). And it looks like advanced summon should give us better self-parking since he hints that advanced summon will do better in parking lots. it also appears that Tesla's alpha development has reached a rough FSD. It can self-drive on local streets and handle intersections and stop signs and traffic lights but not good enough yet. I think Elon is most likely overly optimistic when he says that Tesla will consider its FSD to be safe by the end of the year.

The good news (I am the optimist) is that if the development cars can obey traffic lights and stop signs now then perhaps we will get an update to EAP that can do those two things maybe sometime next year?

It also seems like Musk is taking a simplistic view to what FSD is. Basically, he is looking at the basic driving capabilities that FSD needs to do. So for highway driving, the car needs to stay in the lane, slow with traffic, pass slow traffic, change lanes for a highway change, take an on ramp and take an exit. NOA can technically do all these things so Musk goes "check, we achieved highway FSD". And Musk then looks at what FSD needs to be able to on surface streets. A car needs to stop at stop signs, obey traffic lights, navigate intersections, navigate parking lots. Musk goes "our development cars can do all these things but not reliably so check, we got FSD, we just need to improve reliability and we will be good for regulators."
 
The good news (I am the optimist) is that if the development cars can obey traffic lights and stop signs now then perhaps we will get an update to EAP that can do those two things maybe sometime next year?
Those things have never been promised or even hinted at as features for EAP. They are clearly only going to be part of the Partial Self Drive feature.
 
My service center has checked everything from top to bottom and even replaced the brake booster thinking it might be it. I have just decided to wait for HW3 and see if it’s magically fixed
If you're anticipating a HW3 upgrade that must mean you, like me, have paid for FSD. However...

Listened to earnings call.
No FSD before end of this year, and then to be "approved" by "regulators".

Which means that neither of us will be getting a HW3 upgrade this year either, so you're at least looking at next year, and me being outside North America, probably the year after or even later.