-
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
Sounds like FUD to me
As far as I know the only company that’s had a problem with permits is Uber when they tried to test without one in California and Uber when they ran over a pedestrian in Arizona.
Now, I do think Tesla will run into trouble if they have customers do testing without a permit in California (that’s the FUD I’m spreading).
FUD would be if I was saying Tesla will never get approval. Saying that it is a step that has to be taken, one that Elon has repeatedly mentioned himself, and originally suggested it would take 1 billion miles IRC, is not FUD. It's rational thinking.
Daniel in SD
(supervised)
Well I believe that Tesla will never get approval with HW3 because they will never have a working product but I don’t think that’s FUD, it’s just my opinion. You claimed that the regulatory issue is the primary thing preventing Tesla from meeting their timeline but haven’t really provided any evidence that that’s case. Other companies don’t seem to be having any regulatory problems and many states have regulations allowing autonomous vehicles. Getting the billion miles or whatever Tesla feels is necessary to ensure safety is an engineering problem not a regulatory one. I think 1 billion miles is way beyond what is necessary to prove safety better than a human. Humans get in to accidents about every 150k miles. I bet 10 million miles would be enough but I imagine it’s tricky because you’ve got to account for accident severity not just quantity.
I am quite certain that regulations will not be a problem. We need regulations to counter irresponsible companies like Uber. But once autonomous vehicles have shown themselves to be clearly safer than human drivers, regulatory approval will come rapidly. This is because the great majority of people want safer roads, and because the insurance industry has a lot of lobbying clout. And it's not just Tesla in this game. All the major auto makers are introducing features on the autonomous driving spectrum. So any anti-Tesla sentiment, or political preference for other auto makers will not come in the form of opposition to autonomous cars.
But I don't think Tesla will get regulatory approval to sell a driverless car to the public until they can demonstrate that their technology is mature. And Musk's stated time line of 2020 is ridiculous. I think we might see traffic light and stop sign recognition within the next two or three years. Recognizing when it's unsafe to remain in the center of the lane, or even necessary to veer out of the lane, due to oncoming traffic or pedestrians and bicycles close to the traffic lane will be a much more difficult nut to crack.
I hope I'm wrong and true FSD becomes available this year or next. In the meantime, EAP is the coolest thing ever and makes me a safer driver. (And now my driving is carbon-zero and my miles are free, thanks to the thermonuclear power plant conveniently located a mere 93 million miles from my home.)
EVNow
Well-Known Member
Yes - it is confusing. There are no clear stated standards by Tesla.
In terms of driving it usually it refers to error (or crash) rates. For eg. reported crashes are 4.2 per million miles in US. That is 4.2/1000000 = 99.9996%. So, six 9s is 1 in 1 million. To say FSD is much better than Humans (4x better), a company has to prove it's software can drive a million miles with just one crash. Obviously, to get statistically significant validation, an AV would have to drive a lot more miles so we can calculate the crash rate with some confidence.
Now, when we talk about specific events like intersections, 1 in 1 million I guess is much better than humans.
In some ways I wonder if it was deflection, and buying time.
He destroyed a lot of credibility with all the talk, and promises of FSD. It was never very clear why there was so much of an emphasis on it anyways. It's not like people were clamoring for it, and it's not like competitors were even close.
Did it really help to sell cars? Does it really do that much today than AP1 did before? Not yet. Did it add a ton of liability in terms of promises, and expectations? Yes.
So what actually happened during this time?
They vertically integrated the entire ADAS system in their vehicles, and they're the only manufacture with complete control over it. So they're entirely free to scale it to fit whatever needs they have for it.
Then there is the gigafactory that allows Tesla to build EV's at production rates completely unattainable by other manufactures.
The numerous improvements in efficiency have made it extremely difficult for competitors to match up with range/price let alone the Supercharger network.
There is also the entire Industrial Energy stuff they do that's practically ignored.
They re-positioned themselves as an energy company, and no one really seemed to question the implications.
They're playing around with Tunneling to make special tunnels for Tesla EV's, and scaling up the speed at which autopilot can travel in them.
People are going to look back, and they're going to ask themselves why they missed it. Because Elon promised something so utterly ridiculous that we didn't pay attention to what was really going on. The bad press might seem to do Tesla harm, but I think it's actually acting as camouflage. How not to be taken seriously is to act ridiculous while at the same time doing some very sane things that don't get talked about. The sanity is the incremental improvements.
EVNow
Well-Known Member
EVNow
Well-Known Member
AP/FSD definitely helps Tesla sell cars. Not the promises of tomorrow but today's features - like even the sentry mode.
But here is the thing. AVs are definitely the future (if attainable) - and if Tesla wasn't trying to get there analysts would be asking what they are doing about it. Just look at all other auto majors - they get asked the AV question all the time. Infact most of their boards consider AV the single biggest threat - not EVs, but AVs - with good reason. AVs will probably bankrupt most auto companies because they will likely plummet the car sales - not as many cars would be needed.
Another way of looking at it is - when Netflix moved from mail-in DVDs to streaming, nobody asked for it. There was no competitor. Infact there was a lot of market resistance - the stock even plunged losing 75% of its value.
Netflix Gets Out of DVD Business, Spins of Qwikster
EVNow
Well-Known Member
Musk has not talked clearly about levels. We should really treat FSD as a brand name, rather than a generic term with a specific meaning and attach L5 as the meaning.
Saying "sleep in the car" could be construed as L5. But that is not correct - you could be L4 and within the constrained area sleep in the car. Realistically we should all be talking about L4 - and not L5 at all. Tesla's aim - initially atleast - should be to be at Waymo's level (L4) but in top 200 metro areas of US (and EU). Once they achieve that, then they can talk about L5.
Daniel in SD
(supervised)
Uhmmm. Elon Musk never said parking lots were difficult until very recently. The article is from April.
How disappointing that you continue to misrepresent both Musk’s statement and my position this way.
Elon did not say Level 5 by the end of 2019, nor did I. He said Tesla would be ”Level 5 (no geofence) feature complete” by the end of 2019. My position is simply one of taking Musk for his word.
If something if ”feature complete” — assuming any traditional understanding is followed — all the features for the target outcome have been implemented, but they do not yet have the maturity for general availability. In the case of SAE Levels this obviously can mean prototypes that do not have the reliability to work without a safety driver.
It is very simple. Either Tesla has every feature Level 5 needs implemented by end of 2019, which is what Musk claimed according to the traditional understanding of the words feature complete, or they do not in which case Musk was mistaken.
No, at the Autonomy Investor Day Musk clearly said feature complete by end of 2019 means Level 5 (no geofence) feature complete.
Now of course that means on prototype level, like Uber drives Level 4 prototype cars with safety drivers, not production level without drivers. The Uber’s (as unreliable as they are) are still Level 4 — and Musk’s announcement clearly was Level 5.
Last edited:
No, @diplomat33, you are very mistaken in your interpreration. While it is true ”feature complete” is a software development concept, it is fully compatible with what SAE says about the levels in J3016.
But most importantly you make a huge mistake in how SAE defines reaching its levels. SAE clearly defines Level as design intent, not reliability. If feature complete does not meet Level 5 feature requirements by end of 2019, then Musk was wrong (or worse, lied), because it is fully possible even on a prototype level.
”Levels are assigned, rather than measured, and reflect the design intent for the driving automation system feature as defined by its manufacturer.”
”it is incorrect to classify a level 4 design-intended ADS feature equipped on a test vehicle as level 2 simply because on-road testing requires a test driver to supervise the feature while engaged, and to intervene if necessary to maintain safe operation.”
”An ADS feature designed by its manufacturer to level 5 would not automatically be demoted to level 4 simply by virtue of encountering a particular road on which it is unable to operate the vehicle.”
So it is perfectly possible for Tesla to have Level 5 design-intended prototype vehicles in operation with safety drivers by end of 2019, if their design-intent is ”Level 5 no geofence”. And they certainly can be ”feature complete”, as defined by traditional software development understanding of the word, even if this is not something they could yet ship as Level 5.
Now whether or not Tesla truly is ”feature complete” for Level 5 no geofence features by end of 2019, is of course an open question and we can not know if Musk was truthful or correct in this assessment at Autonomy Investor Day. But he did say it.
By the way, I would also accept a Level 2 design-intent with every feature needed for eventual Level 5 implemented on a prototype level (from Tesla by end of 2019), as feature complete too, since software development definition of the word does not discern between design-intent. But ”Level 5 no geofence feature complete” can not suddenly mean lacking features that a Level 5 car obviously would need.
Last edited:
Basically what Tesla/Musk claimed at the Autonomy Investor Day is: Tesla will have implemented all the features that a Level 5 (no geofence) car needs by end of 2019.
”Level 5” and no ”geofence” is well defined by SAE as a design goal (SAE J3016). (Given that SAE’s definition of Level 5 allows for market limitation, a benign interpretation of Tesla/Musk’s words is within the U.S. of course.)
This can not be disputed, even if we can debate whether or not those features would be required to work or how well (or how well NNs should be trained etc nuance).
Clearly the features to reach SAE Level 5 must exist for it to be feature complete — and that is what Tesla/Musk claimed they’d be by the end of the year.
Neither the traditional definition of ”feature complete” or SAE definition of ”Level 5” require those features to be reliable yet, they can be prototype features that require testing, safety drivers etc.
Level 5 no geofence feature complete, thus, is a perfectly well defined generic goal — if that really is their goal — for the end of the year.
”Feature complete”, as a baseline at least, is well defined by software development tradition. Feature complete - Wikipedia
”Level 5” and no ”geofence” is well defined by SAE as a design goal (SAE J3016). (Given that SAE’s definition of Level 5 allows for market limitation, a benign interpretation of Tesla/Musk’s words is within the U.S. of course.)
This can not be disputed, even if we can debate whether or not those features would be required to work or how well (or how well NNs should be trained etc nuance).
Clearly the features to reach SAE Level 5 must exist for it to be feature complete — and that is what Tesla/Musk claimed they’d be by the end of the year.
Neither the traditional definition of ”feature complete” or SAE definition of ”Level 5” require those features to be reliable yet, they can be prototype features that require testing, safety drivers etc.
Level 5 no geofence feature complete, thus, is a perfectly well defined generic goal — if that really is their goal — for the end of the year.
Last edited:
Let's just be clear. Waymo has a demonstrated, working taxi service in a limited area. Tesla has nothing but Musk's BS and some janky level 2 driving aids.
Sorry to be so blunt, but that's the reality of the situation.
Let's be more clear. On Autonomy day, Tesla took investors on rides in fully autonomous robo-taxis in a limited area.
Limited is limited until it covers your commute.
Were they fully autonomous robotaxis though?
We know they mostly performed the drive autonomously (at least one disengagement happened through driver intervetion) but were they really autonomous in the sense a Level 4 car is? Were they capable of achieving minimal risk condition at any moment for example, without driver intervention?
This is the thing about especially Level 4-5 (but also about Level 3 to a bit more limited extent), it is not just about being able to perform the ”Autopilot thing” well. Autonomy is about a whole lot more.
If you asking whether they (or anyone else's cars) were equivalent to a 16 year old driver, a 22 year old driver, a 35 year old old driver, or a 60 year old driver, I have no idea.
No, I am asking whether their ability to be autonomous existed at all (not its reliability level).
Similar threads
