Elon: "Feature complete for full self driving this year"

[Bladerskb]

I cannot beieve you chose to read his statement in the most optimistic way possible.

What he is really saying: "Tesla will most definitely NOT remove hands on requirement until Oct 2019 AND for unknown time after that"

I read it that way because if he only made the above statement then I would have summed that they were activating their interior cam and using it to do driver check. But then Elon rejected the idea multiple times ot the point that Lex was visibly shocked. Elon's response was that the system is improving so much, so fast, that it will be completely moot, pointless, irrelevant and unsafe and that this will happen very very soon and very very quick.

He said "in fact, I think it will become very, very quickly, maybe and towards the end this year" What will happen towards the end of 2019? "having human intervene will decrease safety. DECREASE!". So he is saying the system will be so good, it will be better than human drivers and towards the end of 2019, having humans intervene (alluding to the question asked of him about adding driver monitoring system for human supervision) would actually make it less safe.

So yes Elon is literally saying they will have L5 fully autonomous in 6-9 month, if not then he would be shocked if it doesn't happen next year.

 

[CarlK]

More verification of how cars in the field train the NN.
Elon Musk on Twitter
"Your interventions do train the NN, but we first verify that the training doesn’t cause other issues, so expect step rather than continuous changes in capability"

One more item that had been debated is clarifird. All AP2 and above cars can be upgraded to FSD.
Elon Musk on Twitter

Anyway we will know a lot more between now and April 22. Those who have been creating their own narratives should stop cooking crows or you will get a bad case of diarrhea.

 

[verygreen]

More verification of how cars in the field train the NN.
Elon Musk on Twitter
"Your interventions do train the NN, but we first verify that the training doesn’t cause other issues, so expect step rather than continuous changes in capability"

Area man known for his exaggerations makes more wild claims that are not supported by independent observations, more at 11!

 

[mongo]

I read it that way because if he only made the above statement then I would have summed that they were activating their interior cam and using it to do driver check. But then Elon rejected the idea multiple times ot the point that Lex was visibly shocked. Elon's response was that the system is improving so much, so fast, that it will be completely moot, pointless, irrelevant and unsafe and that this will happen very very soon and very very quick.

He said "in fact, I think it will become very, very quickly, maybe and towards the end this year" What will happen towards the end of 2019? "having human intervene will decrease safety. DECREASE!". So he is saying the system will be so good, it will be better than human drivers and towards the end of 2019, having humans intervene (alluding to the question asked of him about adding driver monitoring system for human supervision) would actually make it less safe.

So yes Elon is literally saying they will have L5 fully autonomous in 6-9 month, if not then he would be shocked if it doesn't happen next year.

First off, none of that is saying L5 specifically, could be L4.

Second, Elon is taking about skipping L3 entirely. So no need to monitor the driver in a hands off wheel situation.

Third, the end of year timeline is what he has already estimated for feature complete FSD, so that lines up. However, he also said that it would need real world validation to confirm. So likely safer than average driver, L4, but nags until proven & approved. In other words no sleeping while the car drives until 2020 or later.

 

[Randy7fx]

Area man known for his exaggerations makes more wild claims that are not supported by independent observations, more at 11!

What does more at 11 mean? I follow you on Twitter and very interested in what you have to share.

 

[ChrML]

Twitter

Elon is making it very clear, full self driving will be a LOT more expensive than it is today. That is the only way the current vehicles will become "appreciating assets". But they can't increase the price substantially before they roll out more progress. These strategic moves suggest they have recently passed a critical milestone for their HW3 dev builds.

I've said it before, I bought FSD for 3000$ a year ago because if they can deliver any kind of autonomy at L4 or higher, that price was a BARGAIN. Risky buy, yes. It really surprised me that FSD was offered even cheaper for a short period, suggests Tesla really needed money (altough a bit pissing to keep warm).

Most self-driving laws so far say that the vehicle manufacturers are the legal drivers of self-driving cars. This makes it obvious that FSD can only economically be offered by Tesla as a service, or as a really expensive purchase with a capped maximum mileage. The possibility to buy a car with FSD for only 2-5% of the value of the car with infinite mileage is really a once in a lifetime opportunity, because at this point Tesla needs money upfront for R&D. The backside is the risk of Tesla not pulling it off on the hardware as they thought.

A finished feature like promised is probably worth roughly 1500-2500$ annually, or 20000$ as a purchased feature capped at 500 000 km / 10 years. If regular people aren't willing to pay that, leasing and ridesharing service companies will so people won't have to own a Tesla to use a Tesla every day.

 

[diplomat33]

Where in that context did Elon say "when city driving is released and FSD city driving is feature complete"?

They already have FSD and you said it yourself literally acouple pages ago that "there's no point addressing NOA as EAP when Tesla and Elon is now calling it FSD". We literally just had this debate....

This is your interpretation and you are injecting your own ideas and criteria to what he said. This is classic eisegesis. Elon already said in Q4 earnings call that NOA which was "FSD on the highway" was already "feature complete". Do you think NOA will wait on city driving before its performance and reliability improve? Of-course not, we have seen it by the fact that NOA has improved since its original introduction in Sept 2018.

Now if Elon had just said "hands on wheel detection for at-least another 6 months"

Then that to me would mean nothing. As I would believe Tesla was going the Supercruise route and will use the rear-mirror cam for driver check.

But its Elon's second statement that actually meant something. This is where he rejects driver monitoring, his conclusion was that the system is improving so fast, so much, that it will be completely moot, pointless and irrelevant and it will be irrelevant very very soon and very very quick.

Then he gave a timeline when driver supervision and intervention will not be needed. Infact he said that a driver intervening would actually DECREASE safety. So that's clearly a timeline where he believes that the system will be better than a human. Now what was that timeline?

"in fact, I think it will become very, very quickly, maybe and towards the end this year"

This corresponds to needing hands on detecting for atleast another 6 months.
towards end of this year = Q4 2019
hands detection for atleast another 6 months = Oct 2019

Someone saying that Elon portraying that he will have level 5 full autonomy in 6 months is absolutely right. To get technical you can replace 6 months to 6-9 months.

First, NOA is part of FSD but Tesla has not finished releasing all of the FSD features. Maybe NOA nags will be removed in 6 months but Tesla cannot remove all nags for software that is not finished yet.

Second, Musk was specifically asked about L4 so I am not sure where you are getting L5 from. He was asked about L4 specifically and he replied that he thinks FSD will be safer than humans "maybe towards the end of this year" (direct quote) and he thinks that the hands on wheel requirement might be removed "6 months or something" (direct quote). That is not "L5 in exactly 6 months". Sorry. You can't make that leap.

 

[diplomat33]

These strategic moves suggest they have recently passed a critical milestone for their HW3 dev builds.

I think so. All the signs point to it. The April 22 event should shed some light.

 

[CarlK]

Area man known for his exaggerations makes more wild claims that are not supported by independent observations, more at 11!

Known by who? By people who have no idea of how things are done there? Can you list people who have made "independent observastions" and their knowledge and credentials? I'm thinking you're going to get a bad case of diarrhea by 4/22 if not already so.

 

[verygreen]

What does more at 11 mean? I follow you on Twitter and very interested in what you have to share.

More at 11 is just a joke. Urban Dictionary: news at 11 more at eleven - Wiktionary

Known by who? By people who have no idea of how things are done there? Can you list people who have made "independent observastions" and their knowledge and credentials? I'm thinking you're going to get a bad case of diarrhea by 4/22 if not already so.

It is known by everybody who follows Tesla, that Elon Musk makes wildly optimistic statements that rarely (never?) come true on time or match the actual things going on.

The independent observations were performed by a bunch of people, here's a small set: How much data does your HW2+ Tesla upload? you too can perform these observations if you so desire to confirm your car barely uploads anything no matter if you do or do not do any disengagements. But for some reason you appear to be unwilling to perform these observations and letting us know what you saw.

Also please don't poison my food, thanks.

 

[Bladerskb]

First off, none of that is saying L5 specifically, could be L4.

Second, Elon is taking about skipping L3 entirely. So no need to monitor the driver in a hands off wheel situation.

Third, the end of year timeline is what he has already estimated for feature complete FSD, so that lines up. However, he also said that it would need real world validation to confirm. So likely safer than average driver, L4, but nags until proven & approved. In other words no sleeping while the car drives until 2020 or later.

This has nothing to do with L3 nor his previous feature complete by end of the year statement. but keep twisting to absolve him of any responsibility. He literally said human intervention would decree safety towards the end of 2019.

 

[J1mbo]

Pretty clear what is going on... AP is now being given away free. So, the FSD price is going up to make up for the loss of revenue.

The hype at the moment is a cash raise (IMO). Leased M3's going into the Tesla Network starting 2022 / human intervention being less safe than FSD from Oct 2019 / FSD price increase on 1st May ... are you feeling like pulling the trigger yet? If not, we got a show for you on the 22nd...

 

[Bladerskb]

Known by who? By people who have no idea of how things are done there? Can you list people who have made "independent observastions" and their knowledge and credentials? I'm thinking you're going to get a bad case of diarrhea by 4/22 if not already so.

Everyone..

December 2015: "We're going to end up with complete autonomy, and I think we will have complete autonomy in approximately two years."

Elon Musk Says Tesla Vehicles Will Drive Themselves in Two Years

January 2016: "In ~2 years, summon should work anywhere connected by land & not blocked by borders, eg you're in LA and the car is in NY"

Elon Musk on Twitter

June 2016: "I think we are less than two years away from complete autonomy, safer than humans, but regulations should take at least another year," Musk said.

Two years until self-driving cars are on the road – is Elon Musk right?

March 2017: "I think that [you will be able to fall asleep in a tesla] is about two years" -

Transcript of "The future we're building -- and boring"

Nov 15, 2018: "Probably technically be able to [self deliver Teslas to customers doors] in about a year then its up to the regulators"

Elon Musk on Twitter

Jan 30 2019 "We need to be at 99.9999..% We need to be extremely reliable. When do we think it is safe for FSD, probably towards the end of this year then its up to the regulators when they will decide to approve that."

Tesla Q4 Earnings Call

Feb 19 2019 : "I think we will be feature complete full self driving this year. The car will be able to find you in a parking lot, pick you up, take you all the way to your destination without an intervention this year. I'm certain of that. It will be essentially safe to fall asleep and wake up at their destination towards the end of next year"

On the Road to Full Autonomy With Elon Musk — FYI Podcast

April 12th 2019 : "I think it will require detecting hands on wheel for at least six months.... I think this was all really going to be swept, I mean, the system is improving so much, so fast, that this is going to be a moot point very soon. No, in fact, I think it will become very, very quickly, maybe and towards the end this year, but I say, I'd be shocked if not next year, at the latest that having the person, having human intervene will decrease safety. DECREASE! (in response to human supervision and adding driver monitoring system)"

 

[verygreen]

Everyone..

December 2015: "We're going to end up with complete autonomy, and I think we will have complete autonomy in approximately two years."

But if you read my twitter replies, Elon met 100% of his stated goals if we exclude the time component (which they apparently mean that he was not right on time, but he did deliver. Sadly those time travelers don't want to share various stock market statistics for the future years)

 

[dgatwood]

Yes NOA is FSD according to Tesla's new list. But FSD is not feature complete yet since it still lacks the city driving part.

IMO, right now, that's a bit like saying the Model X isn't an airplane because it doesn't have ailerons.

The current stack doesn't even have reliable lane keeping, at least on HW2/2.5, much less city driving. I'm assuming that a lot of the problems will go away when they have enough processing power to throw at the problem, but right now:

• It likes to cross the double-yellow line on tight curves just to see if I'm paying attention. I counted five times yesterday alone before I gave up and drove the rest of the trip by hand.
• It freaks out completely if it sees a turn lane appear simultaneously on both sides (showing the "Take Over Immediately" error).
• When the center line disappears for 50 feet or so, it incorrectly tries to center itself in the "wide lane", and then proceeds to favor the oncoming traffic (left) lane when the center line actually reappears, rather than going on the correct (right) side of the yellow line.

What's interesting is that many of the "wide lane" bugs have gotten objectively worse in the past couple of weeks. In 2018.50.6, I used to kick the thing into AutoSteer mode and let it drive CA-17 nearly all the way, kicking it out only if I was making some tricky lane change. I took over because of bad lane keeping maybe every tenth round trip or so. And even then, it was always in predictable spots.

In the 2019.8.[3,5] release, I'm averaging one serious AutoPilot error PER DIRECTION, PER TRIP, mostly at spots that never caused problems before, even prior to the 9.0 release. (To be fair, it doesn't try to mount the center barriers like it did in pre-9.0 releases, so at least there's that.)

Two weeks ago, I was looking forward to advanced summon, and thinking Tesla might actually be able to pull off something amazing. Now I'm downright terrified at the entire idea of trusting this software stack and the engineering team responsible for it. Such an epic failure of a release should never have been rolled out to the entire fleet. If the full-self-driving feature were available already and its lane keeping got this much worse in a minor software update, I can guarantee there would have been double-digit deaths in the first week.

IMO, Tesla needs to completely reevaluate their bug tracking, crash telemetry, etc. from top to bottom. Whatever they're doing now isn't working. At all. If they don't, we'll continue to have minor improvements in one area of AutoPilot behavior coupled with huge regressions in other areas, and we'll never even approach full self driving. Ever.

A good start would be decoupling the neural network and AutoPilot logic changes from the software updates so that each update can run either the old AP stack or an updated one, and then run both versions in parallel — one on the main CPU/GPU and one on the secondary pair, comparing the differences, and then feeding that data back to Tesla and evaluating offline whether those differences would have made things better, worse, or neither. Then, switch the two for a percentage of users, and at every manual intervention, determine whether the older AP stack would have given different commands that would have averted the manual intervention, and if so, flag that data to use for additional training.

If they had been doing this all along, I suspect that 99% of the regressions that we've seen (from phantom braking to wide lane problems) could have been avoided. That is, after all, the whole point of having double computers.

Then again, it seems likely that they're doing more processing than a single AP computer can handle, which is HW2/HW2.5 will probably not start getting better until they can start ripping out functionality that is FSD-specific and shipping those features only on HW3 (and maybe not even then, though rolling the neural net back to a mid-2018 build would be a good start).

 

[timewasted]

Known by who? By people who have no idea of how things are done there? Can you list people who have made "independent observastions" and their knowledge and credentials? I'm thinking you're going to get a bad case of diarrhea by 4/22 if not already so.

If Elon were to shoot someone in the middle of 5th Avenue in New York, would you still follow him?

Elon is going to be correct one of these days, and maybe THIS time is the one where it ACTUALLY happens. This is the same song and dance he’s been doing for YEARS though (see Blader’s timeline post). Christ, a large part of the online Tesla community acts like battered spouses. Doesn’t matter how many times Elon smacks them around, they still claim that he really loves them. It’s painful to see.

 

[ZeApelido]

Thinking about handcuffing Blader in a model 3 with an FSD route to Bolivia..

 

[mongo]

Everyone..

December 2015: "We're going to end up with complete autonomy, and I think we will have complete autonomy in approximately two years."

Elon Musk Says Tesla Vehicles Will Drive Themselves in Two Years

January 2016: "In ~2 years, summon should work anywhere connected by land & not blocked by borders, eg you're in LA and the car is in NY"

Elon Musk on Twitter

June 2016: "I think we are less than two years away from complete autonomy, safer than humans, but regulations should take at least another year," Musk said.

Two years until self-driving cars are on the road – is Elon Musk right?

March 2017: "I think that [you will be able to fall asleep in a tesla] is about two years" -

Transcript of "The future we're building -- and boring"

Nov 15, 2018: "Probably technically be able to [self deliver Teslas to customers doors] in about a year then its up to the regulators"

Elon Musk on Twitter

Jan 30 2019 "We need to be at 99.9999..% We need to be extremely reliable. When do we think it is safe for FSD, probably towards the end of this year then its up to the regulators when they will decide to approve that."

Tesla Q4 Earnings Call

Feb 19 2019 : "I think we will be feature complete full self driving this year. The car will be able to find you in a parking lot, pick you up, take you all the way to your destination without an intervention this year. I'm certain of that. It will be essentially safe to fall asleep and wake up at their destination towards the end of next year"

On the Road to Full Autonomy With Elon Musk — FYI Podcast

April 12th 2019 : "I think it will require detecting hands on wheel for at least six months.... I think this was all really going to be swept, I mean, the system is improving so much, so fast, that this is going to be a moot point very soon. No, in fact, I think it will become very, very quickly, maybe and towards the end this year, but I say, I'd be shocked if not next year, at the latest that having the person, having human intervene will decrease safety. DECREASE! (in response to human supervision and adding driver monitoring system)"

Why do you continue to use predictions from before the MobilEye split???
Post July 2016, your data shows:
March 2017: about 2019
November 2018: about 2019
Jan 2019: about the end of 2019
Feb 2019: probably 2019, sleep in it around 2020
April 2019: likely better than human this year, deffinatly next year.

This has nothing to do with L3 nor his previous feature complete by end of the year statement. but keep twisting to absolve him of any responsibility. He literally said human intervention would decree safety towards the end of 2019.

Um, yes it does (L3). And your own post (quoted above) states that he said maybe for this year and he'd be shocked if not next year regarding drivers making things less safe. So who is twisting things?

No, in fact, I think it will become very, very quickly, maybe and towards the end this year, but I say, I'd be shocked if not next year, at the latest that having the person, having human intervene will decrease safety. DECREASE! (in response to human supervision and adding driver monitoring system)"

Regarding L3:

I read it that way because if he only made the above statement then I would have summed that they were activating their interior cam and using it to do driver check. But then Elon rejected the idea multiple times ot the point that Lex was visibly shocked.

L2 requires driver attention and continuous monitoring, so wheel torque is a valid check. L3 requires driver attention, but hands off, so alertness system (such as camera) is needed. L4 does not need driver attention, so no monitoring system needed.
If Tesla's FSD software is validated as L2, it can jump to L4 and skip the monitoring needed in an L3 system.

 

[diplomat33]

Everyone..

December 2015: "We're going to end up with complete autonomy, and I think we will have complete autonomy in approximately two years."

Elon Musk Says Tesla Vehicles Will Drive Themselves in Two Years

January 2016: "In ~2 years, summon should work anywhere connected by land & not blocked by borders, eg you're in LA and the car is in NY"

Elon Musk on Twitter

June 2016: "I think we are less than two years away from complete autonomy, safer than humans, but regulations should take at least another year," Musk said.

Two years until self-driving cars are on the road – is Elon Musk right?

March 2017: "I think that [you will be able to fall asleep in a tesla] is about two years" -

Transcript of "The future we're building -- and boring"

Nov 15, 2018: "Probably technically be able to [self deliver Teslas to customers doors] in about a year then its up to the regulators"

Elon Musk on Twitter

Jan 30 2019 "We need to be at 99.9999..% We need to be extremely reliable. When do we think it is safe for FSD, probably towards the end of this year then its up to the regulators when they will decide to approve that."

Tesla Q4 Earnings Call

Feb 19 2019 : "I think we will be feature complete full self driving this year. The car will be able to find you in a parking lot, pick you up, take you all the way to your destination without an intervention this year. I'm certain of that. It will be essentially safe to fall asleep and wake up at their destination towards the end of next year"

On the Road to Full Autonomy With Elon Musk — FYI Podcast

April 12th 2019 : "I think it will require detecting hands on wheel for at least six months.... I think this was all really going to be swept, I mean, the system is improving so much, so fast, that this is going to be a moot point very soon. No, in fact, I think it will become very, very quickly, maybe and towards the end this year, but I say, I'd be shocked if not next year, at the latest that having the person, having human intervene will decrease safety. DECREASE! (in response to human supervision and adding driver monitoring system)"

Interestingly, if Tesla does indeed release feature complete FSD this year as they say, then Musk's more recent predictions (2018 onward), will actually turn out to be pretty close.

 

[dgatwood]

Elon is going to be correct one of these days, and maybe THIS time is the one where it ACTUALLY happens. This is the same song and dance he’s been doing for YEARS though (see Blader’s timeline post).

I think it is inevitable that *eventually* we'll see full self-driving on these vehicles. The only real questions are whether we're talking about one year or 30 and whether Tesla will deliver it before a third party like Waymo sells a retrofit kit. After all, the non-computing parts (fly-by-wire steering, brakes, cameras, RADAR, etc.) are in place, so all that's left is software. (Not that this isn't the hard part, of course, but it is also the part that doesn't require replacing the car.)