-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Elon: "Feature complete for full self driving this year"
- Thread starter dckiwi
- Start date
I would give Tesla $10K today (double the price for the FSD upgrade) if they gave me a double-your-money-back guarantee that within two years I could take a nap in the back seat while my car drove me wherever I wanted to go on the roads of Maui normally and regularly travelled by two-wheel drive passenger cars. That is, I would not require that it take me off-road, or off-pavement, or anywhere that 4WD or AWD is necessary or considered advisable.
Similarly, I'd bet anyone (not over the internet, because how could I collect?) a thousand dollars that within three years from today I will not be able, as an ordinary customer, to buy an upgrade that would allow me to take a nap in the back seat while my car drove me anywhere, as described above.
I'm wondering if anybody here thinks that Tesla will be selling cars that can do that, to ordinary customers, within three years. There's all this talk about "feature complete," but what I think Tesla cannot do in under ten years is sell a sleep-in-the-back car to the general public in the U.S.
Similarly, I'd bet anyone (not over the internet, because how could I collect?) a thousand dollars that within three years from today I will not be able, as an ordinary customer, to buy an upgrade that would allow me to take a nap in the back seat while my car drove me anywhere, as described above.
I'm wondering if anybody here thinks that Tesla will be selling cars that can do that, to ordinary customers, within three years. There's all this talk about "feature complete," but what I think Tesla cannot do in under ten years is sell a sleep-in-the-back car to the general public in the U.S.
Elon tweets today. A million robot taxis by the end of 2020. And looks like more waiting before FSD hardware upgrades start.
Just so we note, Musk is telling a clear untruth there.
”all production in past few months has FSD computer” is not true.
HW2.5 even after April 12th build date
As for ”Yes” answers, his latest ”hot take” is an interesting one. Due to supply & demand once FSD is solved, Musk is now saying consumers have a limited to left to buy a Tesla car.
Tesla will stop selling cars once full self-driving is solved, says Elon Musk - Electrek
Knowing Tesla, end of Q4 means next year at the earliest (SCs are needed for deliveries at the end of a quarter). Knowing Musk, end of Q4 can mean anything from 3 months to 6 years.
What is perhaps not surprising is delaying the costly retrofits to a time when more revenue streams from the China factory, Model S/X facelift etc are expected to come online. Tesla’s pre-sales have always been a bit of a pyramid.
Oh I do think he means Q4/2019.
The whole narrative now revolves around a major step change in FSD around the end of 2019. They will be feature complete for Level 5 no geofence then and in 2020 they will start deploying their robotaxis. So the HW2+ fleet needs the FSD computers from that point onwards. It all makes certain amount of sense from that perspective.
I wouldn’t bet money on it being true of course but that is the narrative I think.
Indeed, that's exactly what he means, and it was pure sarcasm on my part. But like you said - it doesn't make it true. Level 5 robotaxis are many years away...
One reason why I try to be extra careful in making note of what Tesla is claiming, is so that we can later assess whether or not it comes to fruition. Regarding FSD Tesla has set three very clear expectations: 1) It will be Level 5 no geofence (though human supervised during testing stage). 2) Level 5 no geofence will be feature complete end of 2019. 3) Driverless Tesla robotaxis are operating somewhere (presumably in the U.S.) in 2020 — expanding on this they have also mentioned a million robotaxis at end of 2020, though that may have been somewhat more aspirational than the other comments that were clearer plans or even definitive announcements.
This will be very easy to assess in a years or year and halfs time. It will be interesting.
Less interesting but notable previous timelines:
- AP1 parity by end of 2016. Not yet there in July 2019 (no speed-sign recognition).
- FSD option pack differentiating features by April-July 2017 (the infamous 3-6 months). None yet in July 2019.
Both Level 4 and Level 5 can use remote supervision and control, and indeed many of the Level 4 cars operating in the world do, but they can not rely on it (on Level 4 this applies within its ODD) so they must handle everything by themselves and always reach of minimal risk condition (MRC) even without supervision.
Prototypes are different of course. A ”Level 5 feature complete” Tesla would not necessarily be ”Level 5” yet, just feature complete but needing human or remote supervision during testing/maturing.
Well, to be clear, on my part I simply acknowledged that NNs have probably muddied the software development terminology enough that I pondered it might fit if, say, Tesla had every ”classic” software algorithm/feature in place for Level 5 no geofence, and all the NNs they are using for that in place, but the only thing left to do was training those NNs better (and testing)... I guess some might call that feature complete in some sense. I’m not sure if the terminology has truly settled there.
Consider this more musings than an opinion on what could be an acceptable interpretation of what Tesla/Musk meant when they claimed Tesla will be ”Level 5 no geofence” ”feature complete” at the end of 2019. Basically I take that as them saying ”we have software developed everything a Level 5 car needs by end of 2019, now we need to test and improve that”.
As for how much training will the NNs need and what it will be compared to, that is an open question. I was personally just pondering what it would take for a system to be ”Level 5 no geofence” feature complete, but what it would take for it to be reliable is another thing...
Daniel in SD
(supervised)
Well, at least it is logically consistent...
I better order a few more Model 3!
Are you saying Musk is (self-)driving demand?
Daniel in SD
(supervised)
I'm not sure it's working. And it's too bad because the Model 3 is a great car. It's too bad that no one wants sedans anymore. They need to get the Model Y out ASAP.Are you saying Musk is (self-)driving demand?
Personally I think their sales for the time being are OK. The demand does need some driving but it seems to be working. I wish they’d drive it in a different though. Margins are probably not great but not non-existent.
I guess I think their risks are more in the product quality and pre-promise/pre-sale department (and associated risks and costs) than in the immediate demand for the car.
FSD promises may not carry much of a stock price risk (as the market does not seem to believe them) but they do carry cost and risk associated with the features pre-sold.
I noticed this reply today at Elon Musk on Twitter.
To me, that's further evidence that they are likely well behind Waymo and Cruise Automation.
To me, that's further evidence that they are likely well behind Waymo and Cruise Automation.
Last edited:
Male Cow Poo!
Stating that FSD will be solved end of this year/beginning next year when Model Y and even the new Roadster still has to go into production. I don't see the Roadster as a Robotaxi, more a fun car to drive yourself.
Dutchie: Ok, Mr. Disagree, have you looked at the below?
Why testing self-driving cars in SF is challenging but necessary
3 latest videos at Cruise
UPDATE: Disengagement Reports 2018 – Final Results
Testing of Autonomous Vehicles - the reports from Tesla, Cruise and Google/Waymo at Testing of Autonomous Vehicles, recognizing that there is a TON of software engineering talent in the SF Bay Area, where Tesla is headquartered?
Why testing self-driving cars in SF is challenging but necessary
3 latest videos at Cruise
UPDATE: Disengagement Reports 2018 – Final Results
Testing of Autonomous Vehicles - the reports from Tesla, Cruise and Google/Waymo at Testing of Autonomous Vehicles, recognizing that there is a TON of software engineering talent in the SF Bay Area, where Tesla is headquartered?
Last edited:
It does sound to me Tesla is still struggling with perception and not just driving policy which would certainly put them behind Waymo.
For Waymo the problem has not been ”seeing the world” in a long time. Their problems are about predicting other players, what kind of safety margins to use and so forth.
But this is returning to guesswork so perhaps not so interesting. I’ll conclude for now simply stating what I’ve said before that once we see what Tesla means by Level 5 no geofence feature complete at the end of 2019 — on HW3 proper — we’ll know where they stand.
They have set themselves a very amibitious task by end of 2019. It will be interesting to see what comes of it.
For Waymo the problem has not been ”seeing the world” in a long time. Their problems are about predicting other players, what kind of safety margins to use and so forth.
But this is returning to guesswork so perhaps not so interesting. I’ll conclude for now simply stating what I’ve said before that once we see what Tesla means by Level 5 no geofence feature complete at the end of 2019 — on HW3 proper — we’ll know where they stand.
They have set themselves a very amibitious task by end of 2019. It will be interesting to see what comes of it.
I suspect we'll get our first inkling of where things stand when the 'new' enhanced summon appears. If it's still stuck at less than walking speed, can't navigate shopping carts or people then hope dims. Effectively what did this hiatus buy in terms of functionality. If nothing, then yes, I'll start siding more with you on futures.
Aren't they solving a vastly different problem from Waymo? I would think if they were direct competitors city mode NoA would be solved first as driving at distance was not w/i the problem domain. i.e. Waymo may work great driving tired families from one side of Universal Orlando to the other but incapable of picking them up at Orlando International.
Similar threads
