That said, as I said earlier, I appreciate what you are doing @diplomat33 — keep up the great conversation!
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Q3 Earnings Call: Questions about FSD
- Thread starter diplomat33
- Start date
^^ It will be this.
When it comes to software, Tesla release MVP first, then improve it over time.
Unfortunately, their definition of "minimum viable" usually disagrees with the (pre-release) expectations Tesla have set
Saying that, I understand that the AP code already allows for online "railroad" style routing, where the car basically hands "realtime" routing duties to the mothership. I am guessing that this feature would provide the car with the necessary routing to deal with complex situations (timed lanes, complex intersections, localised driving behaviour, traffic avoidance, valid parking spaces, etc.), and only use local processing for situational awareness. (Yes, this is bascially HD maps). So, it could be that we see this and it is measurably better than a smart L2.
My understanding of Elon's answer is that Feature Complete is NOT SAE L5 no geofence. Why did Elon say it is during Autonomy Investory Day? I don't know. If I had to guess, I suspect that Elon was not thinking of the technical definition, but just thinking in the broad sense that FC will have features for the full spectrum of driving (parking lots, city streets and highways) with no geofence, hence L5 no geofence in a very broad, general sense.
After Feature Complete, Tesla will work to add other stuff to make the system reliable in all driving situations. When driver supervision is not required anymore, then it will be L5 no geofence. But FC is essentially a minimum viable product. It's the minimum set of features that cover the basics of parking lot, city and highway driving. FC is not L5 no geofence.
If we forget "L5 no geofence" and just go with Elon's answer and what the website has said, I think we actually do have a better idea of the features in Feature Complete now.
Feature Complete is:
- Smart summon
- Auto park
- Park seek mode
- Auto lane change on city streets and highway
- Navigate on AP on city streets and highway
- On and off ramps
- Traffic lights and stop signs
- Intersections and roundabouts
with the caveat that driver supervision will be required and it will require some driver intervention.
Here is the old FSD description:
"Your Tesla will figure out the optimal route, navigate urban streets (even without lane markings), manage complex intersections with traffic lights, stop signs and roundabouts, and handle densely packed freeways with cars moving at high speed. When you arrive at your destination, simply step out at the entrance and your car will enter park seek mode, automatically search for a spot and park itself. A tap on your phone summons it back to you."
FC is basically this but with driver supervision and some driver interventions. It will not be L5 no geofence. That will come later when Tesla is able to add additional capabilities to remove driver supervision.
I would also point out that the 2016 FSD demo video and the Autonomy Investor Day FSD demo video are probably good indicators of the kinds of routes that they are planning to do in Feature Complete. FC is going from your house, navigating some local roads, handling a few simple intersections, some traffic lights and stop signs, taking an on ramp, navigating a highway for a bit, taking an off ramp, navigating some city streets, traffic lights and stop signs, getting to your parking lot, dropping you off and auto parking, all with driver supervision and some driver interventions. That's FC based on Elon's description. It's not L5.
Frankly, I would respectfully suggest that you drop the "L5 no geofence" part. I think it might be confusing you because you are trying to reconcile things that can't be reconciled. You will note that Elon did not mention L5 or no geofence in any of his answers in the Earnings Call. Just say "feature complete" with the understanding that it is the basic features for parking lot driving, city driving and highway driving with driver supervision.
I put "levels" in quotes because I was simply quoting Elon. He used the word "level", not me. I was not trying to make up a new definition of SAE levels. But I agree that it might be confusing. Earlier, I suggested the new term EL for "Elon Level". LOL. But we can use the term step or stage instead.
There is no confusion on my part. I am just not as cavalier as you about the possibility that Tesla lied on Autonomy Investor Day about targeting Level 5 no geofence feature complete by the end of 2019, so I keep the option included in my analysis.
As @J1mbo pointed out in #22, I mentioned and acknowledge the other option all too well too.
Might not be the best idea to follow Elon’s lead on such things. He tends to obfuscate and conveniently redefine well-defined words, possibly to his PR benefit. See: FSD (2016 vs. 2019), Early Access Program (wide).
Last edited:
You are barking at the wrong tree. I am the one guy here who leaves open the possibility Tesla told the truth when they said on Autonomy Investor Day that they are aiming at Level 5 no geofence feature complete at end of 2019.
@diplomat33, @J1mbo etc seem to have already made up their mind that Tesla were not aiming at Level 5 no geofence feature complete at end of 2019, even though they said so to the investors and the public.
Going with explicitly what Elon Musk has stated what "Feature Complete" means: "Car is able to drive from one's house to work most likely without intervention, so it'll still be supervised… intermediate-speed autonomy, which really just means traffic lights and stop signs."
It sounds like anything after stopping and going at intersections is just general improvements to intermediate-speed autonomy not worth calling out as a feature. This makes some sense in that practically every trip will encounter a traffic light and/or stop sign, and if Tesla is counting number of interventions per trip, addressing this feature will benefit everyone and significantly reduce the number of AutoPilot interventions. Following that reasoning, not everyone deals with roundabouts or sharp curves or double-parked downtown traffic, and these situations overall probably falls into less than half of trips taken allowing for the "most" (greater than 50%) claim. Tesla will likely categorize these types of interventions and focus on them based on how often they happen in the fleet.
It seems likely that navigating from home to work will involve positioning in the appropriate lane to turn through an intersection. Given that Navigate on Autopilot does already handle switching lanes, this isn't really a new feature and just needs to be improved to handle the much shorter distances of switching into the correct lane. And similarly for turning through intersections that often don't have explicit lane lines, AutoPilot already handles unmarked roads (just not allowing activating AutoSteer), so he probably considers driving on these virtual/implicit lane situations as just general improvements and not a separate feature.
Personally, my home to work commute is 2 miles with 5 turns, 2 stop signs, and 1 traffic light on mostly residential streets. I try to use AutoPilot as much as possible, and the only interventions are at the stop signs + red traffic lights and turning to a different street. It seems likely this trip would fall into the "most likely without intervention" category.
It sounds like anything after stopping and going at intersections is just general improvements to intermediate-speed autonomy not worth calling out as a feature. This makes some sense in that practically every trip will encounter a traffic light and/or stop sign, and if Tesla is counting number of interventions per trip, addressing this feature will benefit everyone and significantly reduce the number of AutoPilot interventions. Following that reasoning, not everyone deals with roundabouts or sharp curves or double-parked downtown traffic, and these situations overall probably falls into less than half of trips taken allowing for the "most" (greater than 50%) claim. Tesla will likely categorize these types of interventions and focus on them based on how often they happen in the fleet.
It seems likely that navigating from home to work will involve positioning in the appropriate lane to turn through an intersection. Given that Navigate on Autopilot does already handle switching lanes, this isn't really a new feature and just needs to be improved to handle the much shorter distances of switching into the correct lane. And similarly for turning through intersections that often don't have explicit lane lines, AutoPilot already handles unmarked roads (just not allowing activating AutoSteer), so he probably considers driving on these virtual/implicit lane situations as just general improvements and not a separate feature.
Personally, my home to work commute is 2 miles with 5 turns, 2 stop signs, and 1 traffic light on mostly residential streets. I try to use AutoPilot as much as possible, and the only interventions are at the stop signs + red traffic lights and turning to a different street. It seems likely this trip would fall into the "most likely without intervention" category.
I'm not sure what the point to auto summon and park seeking is with driver intervention. Is the driver supposed to watch the car closely all the time it's moving so it can be stopped manually? To really do that, you'd have to be in an empty, single level lot or keep moving so you could see what the car was doing.
Why bother?
Why bother?
I view the "is the car capable of what I thought it would be capable of" and "do I think I was misled" as legitimate questions that do depend on the person.
For me, the answers to those are yes and no.
Seeing how Tesla actually works, Musks statements yesterday should be helpful. I don't think its helpful to compare those statements to prior statements, but that's just me. I have only had a Model 3 for four months.
But given how the car works, its (1) deliver all the features which would, when they are engaged, allow the car to drive itself, (2) at some point, even though the driver can, and will be prompted to, supervise, the features will not in reality require it, and (3) eventually, the software will allow, and regulators will allow, actual self driving.
Its obvious that they are going to get to (1) first, and its obvious those features will require supervision when released. Its also obvious that (2) will precede (3).
It also follows that as long as supervision is required, the whole liability discussion is sheer speculation, since until (3) the car is not FSD.
I do believe they could release stop lights an stop signs by year end.
For me, the answers to those are yes and no.
Seeing how Tesla actually works, Musks statements yesterday should be helpful. I don't think its helpful to compare those statements to prior statements, but that's just me. I have only had a Model 3 for four months.
But given how the car works, its (1) deliver all the features which would, when they are engaged, allow the car to drive itself, (2) at some point, even though the driver can, and will be prompted to, supervise, the features will not in reality require it, and (3) eventually, the software will allow, and regulators will allow, actual self driving.
Its obvious that they are going to get to (1) first, and its obvious those features will require supervision when released. Its also obvious that (2) will precede (3).
It also follows that as long as supervision is required, the whole liability discussion is sheer speculation, since until (3) the car is not FSD.
I do believe they could release stop lights an stop signs by year end.
That was my bad. I saw something online about the earnings call being Nov 4 and posted it. I was wrong on the date. The earnings call was yesterday, not Nov 4.
I am really not trying to be cavalier about it.
IMO, the earnings call has given us now a perfectly good definition of "L5 feature complete no geofence". it is "no geofenced parking lot driving + no geofenced city driving + no geofenced highway driving + driver supervision + some driver intervention". Do you like that definition?
And to be perfectly honest with you, the reason I say it is not L5 is because I feel like I would sound like a total idiot who does not know what SAE L5 is, if I said it was L5. And I feel like I know enough about the SAE levels not to say something so dumb. Plus, I think it is completely unrealistic to think that Tesla will deliver L5 (with driver supervision) by end of this year. I am just trying to take what I think is a reasonable stance based on everything Tesla has said about FSD.
Your interpretation is a perfectly reasonable way of saying Tesla lied on Autonomy Investor Day when they claimed they were aiming for Level 5 no geofence feature complete at the end of 2019.
All I am saying is that we have not yet seen confirmation of that. I agree it is possible and likely.
boonedocks
MS LR Blk/Blk 19”
My dear is that “Driver Supervision” means hands on the wheel with nags and NOT just an alert when it needs help or sees a problem. The nagging on long commutes (me everyday) is quite annoying. I’ll take Level 3 on highways for a long time before anyone else if they would just get that part working.
I believe you and UBS Colin Langan asking the question on Autonomy Day 3:31:45 misunderstood "feature complete." "Level 5 Feature Complete" is not the same as "Level 5" as Elon Musk pointed out on Autonomy Day and in the Q3 call that the first step, i.e., feature complete, still requires active supervision and potential intervention, which actually makes it Level 2 no different from the current high-speed AutoPilot.
He did adjust his timing of step 2 "no supervision" from a 2020 Q2 guess to "by end 2020," but he has been consistent in believing "feature complete" (which sounds like to him is just stopping for traffic lights and stop signs) could happen by end of this year.
To reach step 2, it would seem that Tesla needs to monitor its data on number of interventions and keep improving its features to reduce that number to some acceptable level.
Yes, that exactly would be the lie part. There is no misunderstanding, both the terms Level 5 and feature complete have widely accepted industry definitions. Elon as an engineer in the field would have been familiar with both as well.
If Tesla claimed at Autonomy Investor Day aiming to be ”Level 5 no geofence feature complete” by end of 2019 (which they did) when in actual fact they were aiming simply for ”Level 2 urban lane keeping with traffic light and stop sign stopping feature complete” (which some of you now assume, probably rightly so), that would have been a lie, as no reasonable observer would have assumed ”Level 5 feature complete” means actually a feature set that much smaller than required for even a prototype (non-reliable) Level 5... and it is not reasonable to assume Elon did not know what he was talking about.
So I really hope it was not a lie.
Is it though?
It gets it's speed limits from OpenStreetMaps.
If you find a speed limit that isn't correct for a given road you simply update OpenStreetMaps. Then wait for it to eventually make it onto your car.
Now sure ideally Tesla would have sign reading, and would have some internal mechanism to update OSM maps themselves so the map data was constantly being updated. But, I wouldn't hold my breath on that happening.
I would even suggest using updating OSM for all the places you park at in order to prepare for smart summon. For when/if it ever arrives in Europe.
Similar threads
