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

[Maitri982]

I've seen mine take a 90° exit from a roundabout in January without following a lead car but no sw since then has managed it, so it may be more luck than anything.
Also he did slightly mount the kerb on exit there, which, if not cambered would have resulted in a bitten rim, as the AP had lost all track of lanes at that point.

NOA is amazing and makes driving much much easier and is way better than anything else on the road. BUT...you must be driving something different than I am if you think it is close to FSD on highways. You do need to pay attention because of the several percentage points it does not handle and you will not know when those are coming.

Also, it does just stupid things. I was driving jersey 95 and it merged onto highway like Pro. It was late and the road was mostly empty and there are like 5 lanes per side. For some reason it then decided it needed to changed lanes to "follow route" even though i was not to exit again for many miles (>10). So it changed lanes then it said it was going to go back to right lane again and it did. The it said again it would change lanes to follow route and this time i cancelled, but it kept asking to do it over and over and over even though I kept cancelling. So this is basic things and it didn't really know what it was doing. This was not dangerous, just not smart.

 

[OPRCE]

I saw CH on the monitor 5cm away and saw OH for Ohio.

LOL, that's funny but a quite unintentional optical effect!

I don't know what level of software you're on or how well it handles your highways.

Am on the latest software, 2019.20.4.2 and NoAP is pretty bad here, e.g. merging into the Autobahn fails on 80% of tries. In light traffic it will fairly smoothly suggest passing cars ok (manual confirmation is obligatory thanks to UN/ECE regulations) but persistently slows way down and gets stuck behind trucks for some reason. And the pièce de résistance is that if a passing manoeuvre does not successfully commence within 5 seconds of activating the blinker, it then aborts completely. So overall it is pretty annoying.

you must be driving something different than I am if you think it is close to FSD on highways.

No, rest assured I never thought that!

 

[diplomat33]

Am on the latest software, 2019.20.4.2 and NoAP is pretty bad here, e.g. merging into the Autobahn fails on 80% of tries. In light traffic it will fairly smoothly suggest passing cars ok (manual confirmation is obligatory thanks to UN/ECE regulations) but persistently slows way down and gets stuck behind trucks for some reason. And the pièce de résistance is that if a passing manoeuvre does not successfully commence within 5 seconds of activating the blinker, it then aborts completely. So overall it is pretty annoying.

NOA is definitely better in the US.

 

[wcorey]

NOA is amazing and makes driving much much easier and is way better than anything else on the road. BUT...you must be driving something different than I am if you think it is close to FSD on highways. You do need to pay attention because of the several percentage points it does not handle and you will not know when those are coming.

Also, it does just stupid things. I was driving jersey 95 and it merged onto highway like Pro. It was late and the road was mostly empty and there are like 5 lanes per side. For some reason it then decided it needed to changed lanes to "follow route" even though i was not to exit again for many miles (>10). So it changed lanes then it said it was going to go back to right lane again and it did. The it said again it would change lanes to follow route and this time i cancelled, but it kept asking to do it over and over and over even though I kept cancelling. So this is basic things and it didn't really know what it was doing. This was not dangerous, just not smart.

Yep, I've noticed that from the initial ride home from delivery. That's why when the release came out where you could optionally let the car do it's thing, I chuckled and said no, ask first. But, I don't have to take corrective action. I just ignore the request unless it is a valid lane change.

 

[wcorey]

Am on the latest software, 2019.20.4.2 and NoAP is pretty bad here, e.g. merging into the Autobahn fails on 80% of tries. In light traffic it will fairly smoothly suggest passing cars ok (manual confirmation is obligatory thanks to UN/ECE regulations) but persistently slows way down and gets stuck behind trucks for some reason. And the pièce de résistance is that if a passing manoeuvre does not successfully commence within 5 seconds of activating the blinker, it then aborts completely. So overall it is pretty annoying.

Well, as much as I've take some semblance of heat the last few days regarding belief we'd see FSD, if only with the requirement to prove to the powers that be that you, as driver, are paying attention, by years end, that was only in regards to the US market. Even when we went to Rotterdam, UK and Paris those roads are not at all like US roads. Given what we receive includes the current training set, I can't imagine how much relevant training there is for the non-US road system. If Tesla thinks millions of training miles in the US will be relevant in Europe my guess is that does reflect some whimsical thinking, IMHO. But deep learning is not my area of expertise so I could be wrong. But I am not, at all, surprised with your assessment. It's a world of difference knowing you reflect a Swiss environmental perception as opposed to an Ohio environmental perception.

 

[EVNow]

NOA is definitely better in the US.

I've only tried NOA a couple of times and in the only freeway interchange I'm yet to successfully use it. The interchange is always backed up - and the car waits too long before deciding to change to the correct lane. They need to change driving policy to use traffic information to change lanes early.

 

[Cirrus MS100D]

Also, it does just stupid things. I was driving jersey 95 and it merged onto highway like Pro. It was late and the road was mostly empty and there are like 5 lanes per side. For some reason it then decided it needed to changed lanes to "follow route" even though i was not to exit again for many miles (>10). So it changed lanes then it said it was going to go back to right lane again and it did. The it said again it would change lanes to follow route and this time i cancelled, but it kept asking to do it over and over and over even though I kept cancelling. So this is basic things and it didn't really know what it was doing. This was not dangerous, just not smart.

I don't do the particular drive often, but I THINK this might be something [GPS?] getting confused about whether you're in the "Cars Only" or "Cars, Trucks, Buses" side of the divide, so NoA keeps thinking you're driving "off the route." I noticed this as recently as Wednesday of last week from NYC to PA and while it kept nagging me to "change lanes to follow route", I purposefully crossed over from the "Cars Only" to "Cars, Trucks, Buses" side and it didn't nag me again for the rest of the ride.

 

[EVNow]

Thanks for that. I suspect that either their numbers or their analysis are off. Robo-taxis might indeed be cheaper than owning a car, since right now using Uber is probably cheaper than owning your own car. But people (myself included) enjoy the convenience of not having to call for a ride. Conventional taxis are not the best comparison, IMO, because the real competition to robo-taxis are Uber and Lyft. These are basically getting almost-free labor, using cars that don't have to have the hardware or software for FSD. So I don't see robo-taxis being any cheaper than Uber. In fact, I see robo-taxis being more expensive than Uber because I'd rather ride in a tobo-taxi than an Uber. It will be safer, therefore can command more money. Obviously, JMO.

Personally, I would love to have Level 3: Driver at the wheel but eyes off the road. I'd pay $5K for that in a heartbeat, and if Tesla wants to call it their FSD package, well, okay. But for me, "full self-driving" means I can take a nap in the back seat. So I don't consider anything below Level 4 to be truly FSD.

Currently Uber/Lyft charge about $2/mile and lose a ton of money (because they have to let the drivers keep most of the money - who are also not making all that much money). By charging just $1/mile, Tesla can reap $92k per robotaxi per year - taking into account all expenses like depreciation, tire, fuel, insurance, remote operator … even parking. The analysts are not off - what is unknown ofcourse is if and when Tesla can ever get to true robotaxi.

 

[wcorey]

Currently Uber/Lyft charge about $2/mile and lose a ton of money (because they have to let the drivers keep most of the money - who are also not making all that much money). By charging just $1/mile, Tesla can reap $92k per robotaxi per year - taking into account all expenses like depreciation, tire, fuel, insurance, remote operator … even parking. The analysts are not off - what is unknown ofcourse is if and when Tesla can ever get to true robotaxi.

View attachment 426838

Billable usage over 12 hrs/day (53%). Now that is laughable unless you're in Manhattan perhaps. On Investor Day talking about regular Tesla owners opting into the Tesla Network while they are at work and clearing $30,000/yr...laughable, unless you are in a major metro environment. I suspect the first thing that'll happen is regular taxi drivers will request a car just so they can knife the upholstery in the back seat. I bought a Tesla that had nice appointment. I, for one, am not inclined to let that car loose on unchaperoned riders. Maybe a used base model.

 

[EVNow]

Billable usage over 12 hrs/day (53%). Now that is laughable unless you're in Manhattan perhaps. On Investor Day talking about regular Tesla owners opting into the Tesla Network while they are at work and clearing $30,000/yr...laughable, unless you are in a major metro environment. I suspect the first thing that'll happen is regular taxi drivers will request a car just so they can knife the upholstery in the back seat. I bought a Tesla that had nice appointment. I, for one, am not inclined to let that car loose on unchaperoned riders. Maybe a used base model.

This is for the robotaxi. Not when putting your personal car on the network (I would definitely not do that !).

Utilization rate is actually conservative - but probably will be true only for world's top 100 metros. Uber / Lyft drivers I talk to they usually cite higher utilization rates. There is some research you can find on this - but from a few years back. Not aware of any recent direct data from Uber (https://www.nber.org/papers/w22083.pdf).

Essentially, since robotaxis can run 24/7, they can just do a lot more miles than a human driver. Tesla can adjust the number of robotaxis available in an area to balance utilization vs wait times.

 

[daniel]

Currently Uber/Lyft charge about $2/mile and lose a ton of money (because they have to let the drivers keep most of the money - who are also not making all that much money). By charging just $1/mile, Tesla can reap $92k per robotaxi per year - taking into account all expenses like depreciation, tire, fuel, insurance, remote operator … even parking. The analysts are not off - what is unknown ofcourse is if and when Tesla can ever get to true robotaxi.

That's interesting, but like the Drake equation, it's full of speculation. Right now we can add the "FSD Package" for $5,000 but that does not mean that the actual cost of a Level 5 car will be $5,000 over the base model. 2¢/mile for electricity assumes you are in Washington State getting hydro, or you have solar, but the estimates didn't include the cost of building that solar or the Powerwalls for refueling all those robotaxis at night. The assumptions about utilization rates seem far-fetched.

In any case, I think the tech is a decade away. I'm aware that some here think we'll see robotaxis in a year or two.

 

[EVNow]

That's interesting, but like the Drake equation, it's full of speculation. Right now we can add the "FSD Package" for $5,000 but that does not mean that the actual cost of a Level 5 car will be $5,000 over the base model. 2¢/mile for electricity assumes you are in Washington State getting hydro, or you have solar, but the estimates didn't include the cost of building that solar or the Powerwalls for refueling all those robotaxis at night. The assumptions about utilization rates seem far-fetched.

The whole point of an open model is you can change the assumptions and get your own numbers ! They are assuming 5 miles / kWh and 10 cents/kWh, which is the US average. I doubled from 0.02c/mile to 0.04c/mile, the profit dropped from 92,050 to 89,530 a year

In any case, I think the tech is a decade away. I'm aware that some here think we'll see robotaxis in a year or two.

5 years. FSD is always five years away.

ps : Waymo basically has level 4 AVs today. They are just geofenced. It is not inconceivable that Tesla can get lvl 4 robotaxis running in top 50 US metros in 3 to 5 years. They can also do it in stages (for eg. have the cars drive empty to the customer, let him drive like a car with ADAS and then slowly go back and park itself when empty).

 

[daniel]

Google says the average is 12¢, but actually I'm surprised it's that low.

By definition a Level 4 car could not operate as a robotaxi since it needs a human in the car. And "letting the customer drive" means it's not a robo-taxi, it's a rental car that delivers itself (except that Level 4 cannot do that!) and would be no use to people who cannot drive.

I've read that Uber drivers lose money if they account for the depreciation of the car. Add $5,000 to the cost of the car, let the owner stay home, and I don't see how you do any better. Uber today is the real-world test. Drivers think they're making money because they're extracting the depreciation value as cash. Cut the price of a ride in half and it's a guaranteed loss. The spreadsheet with its speculative numbers shows a profit. But the actual real-world experiment shows otherwise.

 

[electronblue]

By definition a Level 4 car could not operate as a robotaxi since it needs a human in the car.

This is not correct. Level 4 does not need a human driver, if the robotaxi operates within its design area (the ODD). Level 4 can also employ remote control in case of issues further expanding on this.

Indeed there have been driverless Level 4 cars operating in small pockets of the world already.

 

[Matias]

5 years. FSD is always five years away.

Commercial Fusion reactor has for fifty years been fifty years away. Apparently both have the same escape velocity.

 

[wcorey]

In any case, I think the tech is a decade away. I'm aware that some here think we'll see robotaxis in a year or two.

That would be principally Musk himself. If he is just spouting total nonsensical BS then Tesla should fire him. There is a huge gap between decades away and 6 - 12 months away. Musk's exact quote, which was not simply off the cuff put it as 'feature complete' before years end. Let's say that's accurate, that would put beta (AS, NoA, Summons) a few months later. At the 9 month point any director or VP of engineering should be exceedingly accurate with a pronouncement like that. His quote they underestimated the effort for enhanced summon is nonsensical. The release then pullback of ES tends to indicate a development effort off the rails and, yes, someone should be fired.

 

[daniel]

Musk is insanely optimistic. Thus he makes those kinds of statements. But if he were not so optimistic Tesla probably would not have gotten where it is today. We take the bad with the good. Wise folks ignore Musk's timeline statements.

 

[mongo]

Musk is insanely optimistic. Thus he makes those kinds of statements. But if he were not so optimistic Tesla probably would not have gotten where it is today. We take the bad with the good. Wise folks ignore Musk's timeline statements.

Unless he says yesterday (and perhaps today or tomorrow)

 

[wcorey]

Musk is insanely optimistic. Thus he makes those kinds of statements. But if he were not so optimistic Tesla probably would not have gotten where it is today. We take the bad with the good. Wise folks ignore Musk's timeline statements.

If we were talking about you or me or others on here I'd have no complaints. While Musk is the CEO, effectively he is Chief Technology Officer of a public company. They cannot be talking out of their asses as what they say moves markets. He's already gotten in trouble with the SEC and lost his Chairman status over it.
Where I differ with you is your dismissal, "everybody knows he's full of *sugar*". <- no, not what you said but what you really meant. He's a C-level exec and as such he cannot be blowing bullshit out into the public. Notice you don't routinely hear from May Barra, right? When is says, and this is as much of a direct quote as I can give san suffering through the entire Ark interview. BTW, remember what Ark is. "I'm 100% certain, by the end of this year we'll be feature complete on FSD.... Welllll......", <thinks about what he just said>, "Yes, positively, end of the year".
I've spent decades working with people in Musk's position and they cannot say that unless they know that. I say that because they wouldn't be at that level if they were that incompetent. Consequently, it's either a true statement that should have been in an SEC filing, or it was complete BS and he deliberately said complete BS in which case the SEC needs to get involved again, or...he actually is incompetent in which case the board needs to get involved.

Consequently I, for one, believe him. The alternative calls into serious question the survivability of Tesla. Frankly, there is very little that needs to be added to NoA to make it good on secondary roads. 1) recognize traffic lights, stop signs, yield signs (or just always yield), speed limit signs, construction signs, and checking and turning at an intersection. Then there is managing a parking lot. One thing that happened there is serveral people violated their NDAs and released film of it working. It may well be we aren't hearing anything about it now because they learned from their mistake about what level of confidentiality a non-employee can be responsible for keeping. They could sue a non employee for violating an NDA but for what. People tend to value their employment and, with very rare exception, fulfill the terms of their NDAs.

Of course survivability is as it relates to how high of an item FSD is to them. For those that are infinitely more concerned with 0-6 in under 3 secs, well, FSD is a non-issue. Ditto for those who are passionate about electric. For me, the driving force was FSD.

 

[EVNow]

Consequently I, for one, believe him.

FC is such a nebulous thing - esp., when the feature set itself is not publicly defined, meeting that isn't all that difficult. BTW, as I wrote here before the market does not believe him and has thus assigned no real value to the statement when valuing the company.

I'm more interested in when the City NOA will be released to the fleet - and what the market will do. Of course, city NOA is personally very useful. City NOA should put Tesla on the same level as Waymo (privately valued 3x more than entire Tesla). This should happen sometime next year.