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

[diplomat33]

How does SAE define feature complete? I differentiate feature complete from training complete.

As far I as I know, SAE does not define "feature complete", it only defines the role of the ADS (Automated Driving System) and the role of the driver. Specifically, SAE only looks at the following criteria:
a. Whether the driving automation system performs either the longitudinal or the lateral vehicle motion control sub task of the DDT.
b. Whether the driving automation system performs both the longitudinal and the lateral vehicle motion control sub tasks of the DDT simultaneously.
c. Whether the driving automation system also performs the OEDR subtask of the DDT.
d. Whether the driving automation system also performs DDT fallback.
e. Whether the driving automation system is limited by an ODD.

So questions about features like handling intersections or handling merging traffic are implied in the SAE levels, but not spelled out. For example, SAE defines L4 as "The sustained and 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."

In laymans' terms, a L4 autonomous car must be able to drive itself completely without any human intervention, including handling it's own failures, but the ODD can be limited (for example, geofenced or limited to day driving only).

So what "features" would meet that definition?

The other thing is, I believe Elon Musk, whether one loves him or hates him is far closer to the state of FSD at Tesla than any of us. So the choices are one of two things, he is either with full knowledge and forethought bold face lying to owners, investors, investor advisory firms, and the street at large -- or -- he is making a factually truthful statement.

Or option 3, people are misinterpreting his words maybe?

 

[Daniel in SD]

Googling "Waymo Permission" resulted in this:
Waymo gets permission to pick up California passengers with autonomous cars

TechCrunch confirmed the California Public Utilities Commission (CPUC) gavethe permit to the Alphabet-powered company, granting its employees and their lucky friends and fam to try out the vehicles (limited to the Chrysler Pacifica at the moment, but soon to include the Jaguar I-PACE) within a predetermined area of South Bay near L.A., including the neighbourhoods of Palo Alto, Los Altos Hills, Los Altos, Mountain View and Sunnyvale.

What part of the already existing regulations that explicitly allow autonomous vehicles would prevent them from operating in the entire state? Their problem is not the regulations, it’s that they have a limited operational design domain and even within that domain they don’t think their system is reliable enough yet to operate without a safety driver. There really isn’t any evidence that regulations will be a problem and no one ever seems to be able to cite which regulation they think will be a problem.

 

[Green Pete]

What part of the already existing regulations that explicitly allow autonomous vehicles would prevent them from operating in the entire state? Their problem is not the regulations, it’s that they have a limited operational design domain and even within that domain they don’t think their system is reliable enough yet to operate without a safety driver. There really isn’t any evidence that regulations will be a problem and no one ever seems to be able to cite which regulation they think will be a problem.

The permit requires there to be a safety driver present. Its a robo taxi in name only. To have an actual robo taxi would require a different permit...

 

[diplomat33]

Frankly, after driving today through downtown Indianapolis in rush hour traffic, I am skeptical that FC will be able to handle the complexities. You got 3 or 4 lane avenues where some lanes are turn only, with cars all trying to cut in, pedestrians and cyclists crossing streets everywhere, one way streets, all kinds of different intersections, construction zones etc.. It's a mess that even human drivers sometimes struggle with. Heck, companies like Waymo and Cruise, with more redundant hardware than Tesla, have spent many more years than Tesla on just this one type of complex city driving, and they are just barely there now.

Don't get me wrong, I think Feature Complete will give us the "basic" features for city self-driving. I think FC will be able to do most common intersections and handle traffic lights and stop signs. If you are driving through a town with a simple 4 way intersection, I think FC will be able to handle it really well. But that is a far cry from handling complex city driving like a busy metropolis at rush hour. I think/hope that, after FC is done, that FSD will eventually be able to handle that type of city driving at some point, but I seriously doubt that FC will be able to handle it in 2019.

 

[daniel]

I still don't believe that robo-taxis would be profitable. Uber drivers don't even make back the depreciation on their cars, and that's without paying themselves a wage at all. So we're already not paying the drivers. We're just paying the car owner (who happens to be the driver also) a portion of the depreciation on her/his car. And as long as there are affordable cars, most people, most of the time, prefer the convenience of owning their own car. For 2 or 3 months after I broke my arm I had to use Lyft because the painkillers made it illegal and unsafe for me to drive. And it was a royal PITA!

I think the financial analysis that Musk is using to come up with those profit potential figures assumes that when there are robo-taxis people will use them instead of owning a private car of their own, and be willing to pay taxicab rates. Uber and Lyft will continue to be competitors. If the consideration were robo-taxis vs. taxicabs, there might be a competitive advantage to the driverless car. But Uber put an end to that by undercutting taxis.

As for the claim of a million robo-taxis on the roads by the end of 2020, some other posts upthread clarified that: Musk claims that with the "FSD computer" in all new Teslas being built, plus the ones that will be retrofitted into existing cars, there will be one million cars with the "FSD" computer. It's just plain nonsense to claim that a million FSD-equipped cars equals a million robo-taxis, since very few owners will send their cars out for strangers to ride in, smoke in, and spill food and drinks in. I'd pay for the upgrade if I thought we'd really get FSD, but I would never in a million years let a stranger ride in my car unattended, even with video monitoring.

 

[Daniel in SD]

The permit requires there to be a safety driver present. Its a robo taxi in name only. To have an actual robo taxi would require a different permit...

But they do allow "Driverless AV Passenger Service" (Autonomous Vehicle Carrier Pilot ). Waymo doesn't have a permit for it yet but you haven't provided any evidence that they won't be able to get one. Waymo does have a permit to operate without a safety driver in Arizona (though they don't do that yet as far as I know).
Sorry, I just don't like this FUD about "regulations" being a problem. No one ever seems to be able to cite which regulation is a problem.

 

[diplomat33]

It is also worth noting that feature complete will be different depending on the level of autonomy. For example a L4 highway vehicle would not need traffic light recognition on its feature complete list since there are no traffic lights on interstate highways. But traffic light recognition would definitely need to be on the feature complete list for L5.

 

[EVNow]

Are you saying Musk is (self-)driving demand?

How's your shorting coming along ?

BTW, do you work for a competitor of Tesla (on the self-driving side - not automobile) ?

 

[EVNow]

Hey me too! I never mentioned 1990. As for the other... look closer.

Can you tell me what you mean by that ?

 

[EVNow]

And my stance is that Tesla's feature complete will not be L5, period! It will be designed with the same ODD as L5 but it will NOT be L5. That is what I've been trying to tell you guys!

You can argue "But Elon said L5" all you want but I am telling you reality. FC will not be L5 by the end of this year!

I think that is quite obvious from the FC list - and how many Tesla hasn't even been talking about. I doubt they will get even half the things in that list FC by the end of the year.

But hey, Musk has said he is always optimistic - otherwise he wouldn't have started an EV company.

 

[EVNow]

If you are driving through a town with a simple 4 way intersection, I think FC will be able to handle it really well. But that is a far cry from handling complex city driving like a busy metropolis at rush hour.

According to EM ...

Elon Musk on Twitter

Intersections with complex traffic lights & shopping mall parking lots are the two biggest software challenges. Developer branch already mostly works in these scenarios, but massive effort required to get to 99.9999% safety.
​

Would you guys agree those two scenarios are the most complex ? I guess Waymo / Cruise don't try to handle complex parking lots ?

 

[EVNow]

Or option 3, people are misinterpreting his words maybe?

There has been a lot of discussion on why EM is so very optimistic (to the point of being seemingly untruthful). Some of the explanations are
- SV culture wants to set near impossible goals to drive people to achieve more. A culture where if the team achieves the goal, you have somehow failed.
- Having set very aggressive goals internally, he repeats it outside too - making a miss a certainty. He seems to be slowly learning now - for eg. says Y is slated for mid next year, even though internal goal is earlier.
- He thinks only about how great things can be, instead of thinking about all the risks and how things can go wrong etc

In EM's defense he has got his companies to achieve things that at the outset nobody has seriously tried before.

 

[diplomat33]

According to EM ...

Elon Musk on Twitter

Intersections with complex traffic lights & shopping mall parking lots are the two biggest software challenges. Developer branch already mostly works in these scenarios, but massive effort required to get to 99.9999% safety.​

I think that supports my argument that FC will not be able to handle the most complex city driving since the dev software can handle most intersections but Elon admits handling all intersections at 99.9999% is not there yet.

But I guess it depends on whether Tesla releases FC as a beta requiring driver supervision (because it's not at 99.9999% yet) or waits until they do achieve 99.9999% to release FC. Past experience like how NOA was released, would suggest the first approach. But obviously, if Tesla does wait and only releases FC when it is 99.9999% then that is a different story entirely because FC would be L5 then. Can Tesla really achieve 99.9999% on FC (using AP3) by the end of this year? If so, that would be super amazing. I am skeptical.

 

[wcorey]

Elon admits handling all intersections at 99.9999% is not there yet.

It will never be there. I doubt the average human driver can sustain 99.9999%. All it has to do is best human fatalities/million miles and accidents/million miles.

 

[EVNow]

I think that supports my argument that FC will not be able to handle the most complex city driving since the dev software can handle most intersections but Elon admits handling all intersections at 99.9999% is not there yet.

But I guess it depends on whether Tesla releases FC as a beta requiring driver supervision (because it's not at 99.9999% yet) or waits until they do achieve 99.9999% to release FC. Past experience like how NOA was released, would suggest the first approach. But obviously, if Tesla does wait and only releases FC when it is 99.9999% then that is a different story entirely because FC would be L5 then. Can Tesla really achieve 99.9999% on FC (using AP3) by the end of this year? If so, that would be super amazing. I am skeptical.

His language makes it clear to me that six 9s will come later (after they expend that massive effort). BTW, six 9s is 1 error in 1 Million. Much better than humans.

 

[diplomat33]

His language makes it clear to me that six 9s will come later (after they expend that massive effort). BTW, six 9s is 1 error in 1 Million. Much better than humans.

Yes, which is another reason why FC will NOT be L5. Elon has always been very clear that FC would have all the features but be Beta and require driver supervision at first and the 99.9999% reliability would come later. That means that FC will not be L5 at first but eventually get to L5 later. So the folks who are taking Elon's "Yes" to "FC L5 no geofence" comment and claiming that Elon is now promising FC will be L5 without driver supervision this year, are missing it.

 

[Daniel in SD]

Would you guys agree those two scenarios are the most complex ? I guess Waymo / Cruise don't try to handle complex parking lots ?

They do:
Waymo CEO John Krafcik explains why a parking lot is one of the most difficult environments for a self-driving car
Though this customer says it failed in the Costco parking lot. Not surprising...
Waymo Self-Driving Cars Are Getting Confused by Rain, Passenger Says

 

[Green Pete]

But they do allow "Driverless AV Passenger Service" (Autonomous Vehicle Carrier Pilot ). Waymo doesn't have a permit for it yet but you haven't provided any evidence that they won't be able to get one. Waymo does have a permit to operate without a safety driver in Arizona (though they don't do that yet as far as I know).
Sorry, I just don't like this FUD about "regulations" being a problem. No one ever seems to be able to cite which regulation is a problem.

Im not suggesting it will be in any way problematic. It is however a pre-requisite. California has a lot of incentive to back Tesla from a tax revenue perspective so I find it likely they will push it through. Conversely it might be terribly difficult to get permits in Michigan. Or it could be easy. Nobody knows, it is as much an unknown as is the actual feature completion of FSD. But it is a known unknown. If Tesla throws taxis on the road without permitting they likely open themselves up to huge liability and one accident at fault or not would be devasting. Like driving without insurance.

 

[wcorey]

Elon has always been very clear that FC would have all the features but be Beta and require driver supervision at first and the 99.9999% reliability would come later.

I don't recall him saying that at all. What I did present as the time timestamp in the Ark interview was simply his reference to after mid 2020, whatever that was 'go to sleep' during drive would come the "march of 9s. 99.999% or "5 9's" is standard and considered 'carrier grade'. That means virtually NO failures ever. They will likely never achieve that as human drivers do not come close. It won't happen in our lifetimes.

 

[Daniel in SD]

Im not suggesting it will be in any way problematic. It is however a pre-requisite. California has a lot of incentive to back Tesla from a tax revenue perspective so I find it likely they will push it through. Conversely it might be terribly difficult to get permits in Michigan. Or it could be easy. Nobody knows, it is as much an unknown as is the actual feature completion of FSD. But it is a known unknown. If Tesla throws taxis on the road without permitting they likely open themselves up to huge liability and one accident at fault or not would be devasting. Like driving without insurance.

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 ).