-
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
EVNow
Well-Known Member
Yes - the question is how many of these sub-features are "edge cases" according to Tesla. This is what I mean when I say the feature set is nebulous. I will post the query to Musk on twitter on the off chance he responds.
EAP was absolutely, definitely, one of the essential reasons I bought my Model 3. There were other considerations, but I would not have bought it without EAP. EAP was the feature that pushed my decision over the top.And I bought it for what it could do at the time, not for what Musk promised it would do in the future, because I didn't believe those promises would materialize during the time I expected to own the car.
The Model 3 is, IMO, the best car you can buy today if you don't need to pull a trailer or transport more than 4 people. (5 will fit, but there might be more comfortable options for 5.) Tesla builds great cars that sell themselves. Musk's nutso claims of FSD this year are unnecessary.
Apparently, "feature complete" is a technical term used in software development and can be applied to a program that does not work. It can refer to a program that the developers think has all the components even though they may be buggy and unreliable.
But when a lay person like myself hears "feature complete" I think it refers to the car as a whole, and that one of the "features" the car needs in order to function is software that does what it's supposed to do. It's a case of the common street meaning of a word being different than the technical meaning of the word. Considering that Musk is speaking to the general public when he sends out a tweet, I think it is disingenuous to say something that sounds like "We'll have cars that can go anywhere without a driver," and then claim that you were actually using a technical term that means that your autopilot software will soon be ready to start Alpha testing.
And in both cases, he's talking about what he says will be the situation in half a year or a year and a half, and we all know that nothing at Tesla happens on schedule.
OK, last time. Feature complete is software development term... anecdotally, much like feature not a bug. Musk is in charge of that group. I believe he said in the Ark interview, he was a software guy. That being the case, he speaks like a software guy using software terminology. It means, roughly, the requirements of the features for that version of the software have all been coded. It does not mean, in AI or NN terminology the training sets have been fully vetted. It may well be the development version of the software does know how to do everything on the list of relevant Level 5 functionality yet, due to limitations of the training set, it doesn't execute them as well as they would like. It doesn't mean there aren't bugs. Feature complete in no way implies no supervision. Supervision is both a training issue and a regulatory issue.
This is not saying he wasn't bold faced lying or that his proclamation is accurate to within a day. It means as the CTO and manager of the AI NN team their software will be feature complete by the end of the year. Training set and validation set (addressing disengagements) is a different effort but they programmed for level 5 compliance and those features will be complete by years end.
So with this I believe I will be exiting this thread. This, IMHO, has become a nonsensical conversation.
However, I would strongly encourage all to read the book by Tom Nichols called "The Death of Expertise". It was written a few years back. It is not a software book. It is a book largely about a looming problem in our society these days. Tom Nichols is/was an instructor at the Naval War College in Newport, RI.
diplomat33
Average guy who loves autonomous vehicles
Oh I know this. Thanks. I posted the software definition of "feature complete" several times and tried to explain this very point.
boonedocks
MS LR Blk/Blk 19”
"But when a lay person like myself hears "feature complete" I think it refers to the car as a whole, and that one of the "features" the car needs in order to function is software that does what it's supposed to do. It's a case of the common street meaning of a word being different than the technical meaning of the word. Considering that Musk is speaking to the general public when he sends out a tweet, I think it is disingenuous"
This ^^^^^^^^^
You weren't specifically the target of that post. Reading this thread makes me feel like Bill Murray's character in Groundhog Day.Oh I know this. Thanks. I posted the software definition of "feature complete" several times and tried to explain this very point.
Again, I was not endorsing Musk's statement, just stating level 5 and feature complete are orthogonal. I know very little about training neural networks but do know the training is related to the reliability of it and level 5 specification of without supervision seems to imply degree of reliability.
Interesting quote I came across, "AI is the second best way to do anything". Discuss.
Lasairfion
Member
Well, I was looking forward to Full Self Driving coming sooner than everyone expects it. But since Tesla will be raising the prices of their cars out of consumer reach as soon as Full Autonomy is solved, I'm going to hope it takes a bit longer. Some countries have a year or more delay beyond US launches as it is, to get our hands on the products.
Not to mention that while in urban areas I expect to see full autonomy decimate car ownership figures (they're low enough already in major conurbations such as London for instance), that's not going to be the case for the rural areas. We still need affordable awesome Tesla vehicles within reach.
I hear that the more Enron stock I buy the longer it takes Tesla to reach their goals right?
Not to mention that while in urban areas I expect to see full autonomy decimate car ownership figures (they're low enough already in major conurbations such as London for instance), that's not going to be the case for the rural areas. We still need affordable awesome Tesla vehicles within reach.
I hear that the more Enron stock I buy the longer it takes Tesla to reach their goals right?
diplomat33
Average guy who loves autonomous vehicles
Honestly, I am not too worried about it. I highly doubt that Tesla will suddenly stop making cars or raise prices to $300k next year. That would be a sure fire way of killing Tesla. When Tesla gets a million robotaxis actually deployed and making money, then yes, maybe, it would probably make sense to just make money from the robotaxis and not sell new cars. But Tesla is years away from that.
Cool. I was wondering why you specifically quoted me.
Yes, I feel like it's Groundhog Day too.
But going by the software engineering definition of "feature complete", it's pretty straight forward what Elon meant. Tesla is designing "FSD" to be "L5 no geofence" system since it will not be geofenced or ODD limited in any way and when finished, will self-drive everywhere with no driver intervention. "feature complete" is the pre-beta software when all the essential features are implemented but the software is not finished yet. For Tesla, "feature complete" is the list of features on the FSD order page. When Tesla gets to "feature complete L5 no geofence", it will mean that they have all the FSD features on the website implemented with no geofencing or ODD restrictions, but not completely finished yet.
It doesn't!
As a layperson you know nothing of the software development life cycle. Spoiler alert, nobody expects you to.
Read "Death of Expertise".
This should be it's own thread...I suspect it will as that stream of consciousness on perhaps the future of Tesla is interesting. If a Tesla can truly, in the right geographic area generate $30,000/yr for it's owner I think that obviously would raise the value proposition of owning one. I live in a rather rural area of CT, between Hartford, CT and Providence, RI. I am seriously thinking of buying a second stripped down Model 3, pull the steering wheel and send it into either city for the day to earn it's keep. It'll be a business expense, depreciated on taxes. and, essentially, free money. What's not to like? Is it possible Tesla will make more money selling taxis than selling personal transportation? I don't know the answer but I know people who would be capable of finding out. Then it's simply supply and demand curve to set optimal pricing. That is right up Linear Programming's alley.
To summarize, I see three main interpretations of ”Level 5 no geofence feature complete” (the Musk comment from Autonomy Investor Day) in this thread:
1) @diplomat33 believes it means some set of AP/FSD features of Tesla’s choice that may not have anything to do with ”Level 5 no geofence” as defined by SAE J3016
2) @electronblue (me) believes it means a full set of features required by Level 5 no geofence on prototype level, but the features may be unreliable and alpha and can require safety driver (as defined eg in Wikipedia https://wikipedia.org/wiki/Feature_complete and SAE J3016 respectively)
3) Some others feel ”feature complete” means not only a full set of features required by Level 5 no geofence (as defined by SAE J3016), but those features must also be at least somewhat reliable (some legitimate discussion about different definitions of feature complete within software development)
And nobody seems to believe Musk.
Did I capture the sentiment about right?
1) @diplomat33 believes it means some set of AP/FSD features of Tesla’s choice that may not have anything to do with ”Level 5 no geofence” as defined by SAE J3016
2) @electronblue (me) believes it means a full set of features required by Level 5 no geofence on prototype level, but the features may be unreliable and alpha and can require safety driver (as defined eg in Wikipedia https://wikipedia.org/wiki/Feature_complete and SAE J3016 respectively)
3) Some others feel ”feature complete” means not only a full set of features required by Level 5 no geofence (as defined by SAE J3016), but those features must also be at least somewhat reliable (some legitimate discussion about different definitions of feature complete within software development)
And nobody seems to believe Musk.
Did I capture the sentiment about right?
diplomat33
Average guy who loves autonomous vehicles
To summarize, I see three main interpretations of ”Level 5 no geofence feature complete” (the Musk comment from Autonomy Investor Day) in this thread:
1) @diplomat33 believes it means some set of AP/FSD features of Tesla’s choice that may not have anything to do with ”Level 5 no geofence” as defined by SAE J3016
2) @electronblue (me) believes it means a full set of features required by Level 5 no geofence on prototype level, but the features may be unreliable and alpha and can require safety driver (as defined eg in Wikipedia https://wikipedia.org/wiki/Feature_complete and SAE J3016 respectively)
3) Some others feel ”feature complete” means not only a full set of features required by Level 5 no geofence (as defined by SAE J3016), but those features must also be at least somewhat reliable (some legitimate discussion about different definitions of feature complete within software development)
And nobody seems to believe Musk.
Did I capture the sentiment about right?
Not quite. I believe "feature complete L5 no geofence" is the set of FSD features specifically mentioned on the FSD website that will have the same ODD as L5 as defined by the SAE. It will be a prototype and will require a safety driver at first."
So basically you are cherry picking the ODD from Level 5 but not any of its other requirements?
No, I tend to believe Musk. And, though I am not speaking for him, it appears so does @Green Pete.
To disbelieve him implies deliberate misinforming the market. To believe he doesn't understand is unabashed ignorance of what power a C-level exec has and what information he is privy to.
To infer uninformed opinion trumps experienced based assessment is precisely the topic of "Death of Expertise", which I encourage all to read.
Reflecting on observations expressed in the book with the backdrop of this conversation is eerie to say the least. And the book is like 2 yrs old or so.
The problem is, @wcorey, there is also a logical fallacy called ”appeal to authority”. Sometimes the authority is also wrong.
It is hard to tell which applies and when, especially since C-level executives also have an inherent bias: they work for a business they are trying to advance. They are not impartial observers by any means and thus have their own agenda (as they should). Observers should realize this and take it into consideration.
In the end we don’t know but we can and should continue to observe critically and adjust as necessary.
It is hard to tell which applies and when, especially since C-level executives also have an inherent bias: they work for a business they are trying to advance. They are not impartial observers by any means and thus have their own agenda (as they should). Observers should realize this and take it into consideration.
In the end we don’t know but we can and should continue to observe critically and adjust as necessary.
We absolutely should observe critically. But in situations like these where companies do not give the general public access to all the information there are only specific ways you can be critical.
Experts in the field seem to agree, Tesla is far ahead of everyone else because they adapted early in ways that competition has not event done yet, such as putting all the required equipment on production cars years before it had a consumer facing purpose. Collecting the data generated from that hardware to improve the software through Deep Learning.
My assumption is that we are all on the same page for the above.
How close Tesla is to solving that problem is the question. For the most part we are all just spit balling our opinions because the people who know how far Tesla is in the development process are all bound by very powerful NDAs. The things that those people have been able to say seem very optimistic, maybe not quite as optimistic as Musk, but seem to indicate that 1 year later than Musks timeline is likely a very safe bet. Some of these people's entire job is to survey the landscape of things like this and make multi-million dollar bets based on their assessment. Others are just fan-people who got the chance to ride in the car at the Autonomy day demo.
Pretty much anyone not in the group above is just making noise. I am willing to believe Musk in this case because he is looking at data and numbers that nobody else who has spoken publicly has seen. And even if he's wrong the outside analysts with any insight don't put him very far behind what he is saying.
So maybe FC FSD happens sometime next year. I would be disappointed but not surprised. I still think there's a good chance they nail it this year. But to the people saying its 10 years out with zero insight into the actual state of development. You do realize you are making wild baseless claims that someone else who has orders of magnitude more knowledge than you is making baseless claims? Do you not see the irony?
diplomat33
Average guy who loves autonomous vehicles
Correct. The reason I am doing that is because "feature complete" will be a prototype as you said. It will make mistakes. It will require driver interventions from time to time. I am looking at the reliability of the "feature complete L5" and whether it can meet the SAE L5 definition:
The sustained and unconditional (i.e., not ODD-specific) performance by an ADS of the entire DDT and DDT fallback without any expectation that a user will respond to a request to intervene.
The way I see it, "feature complete L5" as a prototype that requires driver interventions will not be able to meet that entire definition of SAE L5 in the beginning. It will as it gets better and more reliable but not in the beginning. "Feature Complete L5" may be designed to be L5 but it won't be L5 in practice since it is a prototype.
But if I am understanding you correctly, you think that "feature complete L5" will have all the features such that it can meet that definition of L5 in its design, just not in reliability? Is that right?
Personally I find these people to be rather a minority. Most of us freely admit we don’t know how long it will take for Tesla to reach Level 4, let alone Level 5. I for one am observing what I can and trying to see behind the PR.
I have a much better guess when some other companies will reach Level 4 though, since we know a few companies have operated driverless cars in limited fashion.
Last edited:
If what Musk was saying was really ”FSD no geofence feature complete”, he should have said that, instead of yes to ”Level 5 no geofence feature complete”. That would have allowed Tesla to re-define what it all (FSD) means. Referring to Design Studio then would have been much more reasonable. But he said Level 5 and that means a not just reliability but a specific set of features implemented.
And let’s not be naive, he knew what he was doing when he said yes to Level 5. He has been waving the Level 5 card since October 2016 when he announced AP2 as Level 5 capable hardware.
SAE J3016 specifically allows for Level 4/5 prototypes that are Level 4/5 even without the reliability to be autonomous so that is no excuse.
Yes, obviously. That is what feature complete and Level 5 mean by their dictionary and SAE J3016 definitions respectively.
Similar threads
