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

[electronblue]

Yes, in Lex's interview Musk agreed with Lex when he said FSD is NOA on freeway + City NOA. They are clearly now working on city features (as Musk has talked about and we see features added to the prod code but not enabled).

But I hope, they have as the feature list to complete the NHTSA list. That will do 2 things
- Legitimately claim FSD complete (and recognize FSD revenue)
- Make them comparable to Waymo

You say this as a good thing but in my view this is actually a concern.

What was sold to us in 2016 as FSD was definitely not just ”NOA on freeway + City NOA” but I can see that Musk and Tesla may be trying to retroactively position it that way now to call it mission accomplished.

What was sold as FSD to us was sleep in your car as well as Summon your car from coast to coast without a driver and to achieve that — and to achieve comparability to Waymo ie car responsible driving — Tesla would actually have to complete of a lot more than just ”NOA on freeway + City NOA”. They would need at least Level 4 for that and in fact Level 5 is what Musk has been projecting both in 2016 (AP2 is Level 5 capable hardware) and 2019 (Level 5 no geofence feature complete at the end of 2019).

So let’s hope FSD is not just ”NOA on freeway + City NOA” in reality and that the 2019 feature complete still means what Musk said at the Autonomy Investor Day: Level 5 no geofence. That is so much more than ”NOA on freeway + City NOA”.

 

[diplomat33]

You say this as a good thing but in my view this is actually a concern.

What was sold to us in 2016 as FSD was definitely not just ”NOA on freeway + City NOA” but I can see that Musk and Tesla may be trying to retroactively position it that way now to call it mission accomplished.

What was sold as FSD to us was sleep in your car as well as Summon your car from coast to coast without a driver and to achieve that — and to achieve comparability to Waymo ie car responsible driving — Tesla would actually have to complete of a lot more than just ”NOA on freeway + City NOA”. They would need at least Level 4 for that and in fact Level 5 is what Musk has been projecting both in 2016 (AP2 is Level 5 capable hardware) and 2019 (Level 5 no geofence feature complete at the end of 2019).

So let’s hope FSD is not just ”NOA on freeway + City NOA” in reality and that the 2019 feature complete still means what Musk said at the Autonomy Investor Day: Level 5 no geofence. That is so much more than ”NOA on freeway + City NOA”.

"NOA on freeway + City NOA" is just the beginning of FSD. It is not the end. FSD will still be L4/5 no geofence when it is finished. "Feature complete" is just the beginning of FSD too, not the end. So FSD will be a lot more than just "NOA on Highway + City NOA".

Tesla uses NOA to link Autopilot with Navigation. So "NOA on Highway + City NOA" allows Autopilot to work on both highways and city streets so it is an important prerequisite for FSD but it is not the end all be all of FSD.

Basically, FSD will come in 2 stages:

Stage 1: NOA on highway + City NOA = "Feature Complete"

Stage 2: L4/5 no geofence

 

[wcorey]

Interestingly enough Tesla is using Agile development, and "feature complete" is something that Agile development eliminates.

Not as such. The term may be obsoleted but the intent isn't. Product will come up with a MVP list and over the sprint cycle features on the MVP are marked completed on the BDL.

Based on the (predictable) problems Tesla have experienced in just getting the narrow subsets of FSD to work properly, they are a long way from having full, proper FSD. Probably 2 to 3 years.

I sure hope not, Musk was adement FSD would be feature complete by end of 2019. If you listen to the Ark interview his comments on this weren't 'off the cuff' they were thought out and carefully chosen/worded.

Note: The delayed rollout of Advanced Summon is an excellent example of the problems trying to get a narrow FSD subset working.
Also the inconsistent performance of NoA's lane change recommendations is another example of a narrow but critical part of FSD not working well.

I suspect what happened with Enhanced Summon was incomplete feature definition. From early video demonstrating features of FSD, dropping off at mall and self parking and retrieval at curb side was much different than follow the bluetooth signal, which is what they delivered. A car in a parking lot can't crawl, it's got to move with deliberate intent. What concerns me though is Musk's comment on 'it was more difficult than we thought'. WHAT????? come on - I been is software development over 45 yrs now. I could have defined a MVF list for enhanced summon. I'd bet on you being able to as well.
For those unfamiliar with the terminology -
MVP is Minimum Viable Product
MVF is Minimum Viable Feature
One could swap out the use of the word feature for the newer 'story'. Without getting bogged down in a word salad terminology game in an Agile setting sometimes expediency wins over purity.

 

[wcorey]

So let’s hope FSD is not just ”NOA on freeway + City NOA” in reality and that the 2019 feature complete still means what Musk said at the Autonomy Investor Day: Level 5 no geofence. That is so much more than ”NOA on freeway + City NOA”.

In the Ark interview, I don't recall clearly the Lex interview, Musk did say go into the back seat and take a nap or read a book. He, further, said 2020 would see the Tesla Network of robo-taxis.

 

[diplomat33]

Tesla has identified 4 domains of FSD and 4 Autopilot Features to handle each domain:

Those are the 4 domains that make up any driving that a person does. You need these 4 domains in order to self-drive from any A to any B.

So "Feature Complete", as Tesla defines it, means getting the car to be able to handle those 4 domains reasonably well. It won't be perfect at first but it will be good enough for the public with driver supervision.

Then, with fleet data, Tesla can improve the features until everything works so well that the driver is not required at all.

In that list, NOA on Highways and Auto Park have been released to the public. So Tesla just needs to release Enhanced Summon and NOA on City Streets in order to achieve "Feature Complete".

But I highly doubt that Tesla will just release the same capabilities of the current NOA (following nav and auto lane changes) but on city streets and call FSD finished and done. "Feature Complete" is just the beginning of FSD, not the end. Once Tesla is "Feature Complete" then they need to improve everything until the driver is not needed anymore.

So the plan is to release a reasonable version of Enhanced Summon and City NOA to the fleet by the end of 2019 and then next year, refine, add, tweak and improve everything until it it so good that the driver is not needed anymore.

 

[OPRCE]

Tesla has identified 4 domains of FSD and 4 Autopilot Features to handle each domain:

View attachment 424685

Those are the 4 domains that make up any driving that a person does. You need these 4 domains in order to self-drive from any A to any B.

So "Feature Complete", as Tesla defines it, means getting the car to be able to handle those 4 domains reasonably well. It won't be perfect at first but it will be good enough for the public with driver supervision.

Then, with fleet data, Tesla can improve the features until everything works so well that the driver is not required at all.

In that list, NOA on Highways and Auto Park have been released to the public. So Tesla just needs to release Enhanced Summon and NOA on City Streets in order to achieve "Feature Complete".

But I highly doubt that Tesla will just release the same capabilities of the current NOA (following nav and auto lane changes) but on city streets and call FSD finished and done. "Feature Complete" is just the beginning of FSD, not the end. Once Tesla is "Feature Complete" then they need to improve everything until the driver is not needed anymore.

So the plan is to release a reasonable version of Enhanced Summon and City NOA to the fleet by the end of 2019 and then next year, refine, add, tweak and improve everything until it it so good that the driver is not needed anymore.

In theory, yes, but in practise Tesla time-dilation is very much a thing ... e.g. Highway NoAP has been out 8 months in USA and is still nowhere near being able to manage without constant human supervision. I expect/hope that in about another year on HW3 it should qualify as a robust and above all reliable Highway L3, but that still leaves a couple more years of work to get to a general purpose L4 vehicle on HW4.

I think Tesla has (compared to Waymo) a relatively small team (of ~250) dedicated to the FSD/AP development, so progress will be much less rapid than in the blessed fantasy of Elon, where all things operate in ideal conditions on Sugarcandy Mountain.

 

[wcorey]

Tesla has identified 4 domains of FSD and 4 Autopilot Features to handle each domain:

View attachment 424685

Those are the 4 domains that make up any driving that a person does. You need these 4 domains in order to self-drive from any A to any B.

So "Feature Complete", as Tesla defines it, means getting the car to be able to handle those 4 domains reasonably well. It won't be perfect at first but it will be good enough for the public with driver supervision.

Then, with fleet data, Tesla can improve the features until everything works so well that the driver is not required at all.

In that list, NOA on Highways and Auto Park have been released to the public. So Tesla just needs to release Enhanced Summon and NOA on City Streets in order to achieve "Feature Complete".

But I highly doubt that Tesla will just release the same capabilities of the current NOA (following nav and auto lane changes) but on city streets and call FSD finished and done. "Feature Complete" is just the beginning of FSD, not the end. Once Tesla is "Feature Complete" then they need to improve everything until the driver is not needed anymore.

So the plan is to release a reasonable version of Enhanced Summon and City NOA to the fleet by the end of 2019 and then next year, refine, add, tweak and improve everything until it it so good that the driver is not needed anymore.

But wait...
Auto Park has a serious flaw. It determines a valid space to be between two parked vehicles not either an edge or a side of an empty rectangle.
______________
| | | |
| | | |
| | | |
_______________

or
_________________
| | |

For myself, I would not consider that feature/story complete unless one could stipulate farthest away no surrounding no immediately adjacent vehicles. Presumably 'come get me' summon would be working. This akin to the story drop me off at the mall then go park. When I signal come back to where you dropped me off. And it will have to handle vertical parking lots as well.. multi-floor parking garages.

 

[wcorey]

e.g. Highway NoAP has been out 8 months in USA and is still nowhere near being able to manage without constant human supervision.

Really? I have virtually no problem with it on interstates and other limited access roads. Several people I know of that traveled cross country said the car literally did 99% of the driving. One thing I will say To get the the nearest theater there is a 19 mile interstate trip. The default behavior of NoA is to not be in the slow (entrance/exit) lane so the first thing it tries is to change lanes. Unfortunately I-395 is a two lane road so unless I want to do 10 over speed limit I have to stay in slow (entry/exit) lane. That's easy enough to do by deselecting auto lane change.

 

[wcorey]

"Feature Complete" is just the beginning of FSD, not the end.

I really hope that's true, as it relates to when OTA updates stop. I have a radio from FlexRadio Systems that do over the internet updates upon request. The issue there is they are a small company and software is expensive to produce so they have to sell product to stay in business. At some point, hopefully not in my life time, they will want you to either pay for another year of updates or buy a new radio. So, the question remains at what point will Tesla declare what we paid for, by way of defining done, is done and if we want new features we need to buy a new car from them.

 

[OPRCE]

Really? I have virtually no problem with it on interstates and other limited access roads. Several people I know of that traveled cross country said the car literally did 99% of the driving. One thing I will say To get the the nearest theater there is a 19 mile interstate trip. The default behavior of NoA is to not be in the slow (entrance/exit) lane so the first thing it tries is to change lanes. Unfortunately I-395 is a two lane road so unless I want to do 10 over speed limit I have to stay in slow (entry/exit) lane. That's easy enough to do by deselecting auto lane change.

Yes, if one cannot set the speed to 90mph and safely focus on a Harry Potter movie until alerted to resume control several seconds later, then it is still not Highway L3, nor do I believe it is 99% of the way there.
That's not to say NoAP is not of some use at the present time but we are talking timescales of its evolution here.

Personally I'd prefer to see Tesla nail down the system safety on the motorway before piling on more features/domains, by demonstrating that they have 3D-vision depth-mapping (pseudo-LiDAR) working reliably enough to enable controlled braking from any speed for a stalled vehicle suddenly appearing in my lane.

If that worked properly then getting qualified to Highway L3 should be a cinch, and a huge tangible advance for users, then on the basis of maximum safety move on to develop in those areas with higher hazard for contact with pedestrians, cyclists and oncoming traffic, etc.

However, as I know Tesla, they will probably do it the other way around, with a maximum of flashy claims but low safety/reliability mentioned only in the small-print, hence likely making "FSD" remain a L2 system for a couple of years after the first "feature-complete" release.

 

[diplomat33]

Yes, if one cannot set the speed to 90mph and safely focus on a Harry Potter movie until alerted to resume control several seconds later it is still not Highway L3, nor do I believe it is 99% of the way there.
That's not to say NoAP is not of some use at the present time but we are talking timescales of its evolution here.

I see you live in Bern. I know in Germany there are highways with no speed limit at all. But here in the US, there are few highways where you are allowed to go 90 mph. In the US, having the car self-drive at 90 mph while the driver watches a harry potter movie is an unnecessarily high standard. You set the bar way higher than I do. I know on the highways I drive, NOA is already close to FSD on the highway. For me, if the car could safely handle highway driving at 70 mph (the speed limit where I live) while I watch a movie and notify when I reach my exit, I would be happy and consider it NOA to be true FSD Highway.

But your post does illustrate the challenge that Tesla has in releasing full self-driving because expectations and standards will vary greatly from location to another. That's why we have folks who gush that NOA is almost FSD while others lament that is still very far from FSD.

 

[OPRCE]

I see you live in Bern. I know in Germany there are highways with no speed limit at all. But here in the US, there are few highways where you are allowed to go 90 mph. In the US, having the car self-drive at 90 mph while the driver watches a harry potter movie is an unnecessarily high standard. You set the bar way higher than I do. I know on the highways I drive, NOA is already close to FSD on the highway. For me, if the car could safely handle highway driving at 70 mph (the speed limit where I live) while I watch a movie and notify when I reach my exit, I would be happy and consider it NOA to be true FSD Highway.

But your post does illustrate the challenge that Tesla has in releasing full self-driving because expectations and standards will vary greatly from location to another. That's why we have folks who gush that NOA is almost FSD while others lament that is still very far from FSD.

Personally I generally toodle along at 125kmh (77mph) on the Autobahn, even when in Germany, where people doing 250kmh (155mph) in the third lane is a regular thing, but Tesla does enable AP up to 150kmh (93mph) over here so I imagine FSD should work up to the same limit.

I think though my proposal is if anything perhaps a shade too low a standard to set for the type of tests measuring safety, as one ideally wants to not aim for the bare minimum but leave some leeway in there, i.e. it should work equally well at let's say 10% over the maximum programable speed.

It seems likely and recommendable that a battery of such standard hazard-scenario tests will be developed in (at least) Europe in the near future, passing which will become a prerequisite to issuing any >=L3 approvals for everyday use. IMHO there is simply not a snowball's chance in Hell that mere statistics (even when heavily sprinkled with Musk's special pixie-dust) are going to sufficiently impress regulators this side of the pond.

 

[wcorey]

Yes, if one cannot set the speed to 90mph and safely focus on a Harry Potter movie until alerted to resume control several seconds later, then it is still not Highway L3, nor do I believe it is 99% of the way there.

A Level 3 vehicle is capable of taking full control and operating during select parts of a journey when certain operating conditions are met. For example, a vehicle that is capable of managing itself on a freeway journey, excluding on- and off-ramps and city driving, might be considered Level 3 automated.Mar 29, 2018
It was 99% of the driving. Level 3 includes city driving. There have been people who tied a 1lb dumbell to the steering wheel. Note "during select parts of a journey when certain other conditions are met". If one is on I-80 cross country from NY (west of NYC) to Ohio, yeah, short of SC stops, I am totally convinced that is doable or I-95 north of DC to I-495 to I-95 south of DC...I'd put money on that being doable as well. I-290 in Mass to I-95 in southern CT non-rush hour, yep.

I didn't read the other post talking KM etc. I saw CH on the monitor 5cm away and saw OH for Ohio. I don't know what level of software you're on or how well it handles your highways.

 

[OPRCE]

It was 99% of the driving

I don't dispute that they did 99% of their cross-country journey on AP, but that is different from claiming the current system has 99% of the capabilities needed to qualify it as a Highway L3 and adding weights to the wheel will not accelerate progress, except for a one-way trip to the morgue, since AP still formally requires constant human supervision whether people like to acknowledge this fact or not.

 

[diplomat33]

Autopilot doing better in roundabouts:

 

[OPRCE]

Autopilot doing better in roundabouts:

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.

 

[AlanSubie4Life]

Autopilot doing better in roundabouts:

I don’t even know what this video means. I’d like to see the car do that “successfully” 10 times in a row. I doubt it would happen. I believe the result could easily be random. Fairly sure no one would expect Autosteer to be able to handle this case.

 

[wcorey]

I don't dispute that they did 99% of their cross-country journey on AP, but that is different from claiming the current system has 99% of the capabilities needed to qualify it as a Highway L3 and adding weights to the wheel will not accelerate progress, except for a one-way trip to the morgue, since AP still formally requires constant human supervision whether people like to acknowledge this fact or not.

I merely mentioned the weight, not endorsing it, as some people trust it to that extent and DO want to watch Harry Potter movies. Please explain how. The definition I gave from Google said nothing about 99%. Is that an EU thing? What I gave you as first hand is accurate. I think when they get approval to go GA, not beta, it'll be level 4 if not 5. They are planning on a robo-taxi fleet next year. Am I buying stock in anticipation? No. Again, you're talking about the car's performance in Switzerland, not US. It may well lag.

 

[Tanc]

Anyone buying the FSD option is essentially foolish. It has no predicted delivery date and only vague statements about its capabilities. The only rational reason to buy it is on the assumption that 1) it will be released someday and 2) the future cost of activation will be significantly higher than the present cost. Without a significant pricing differential you are losing the utility of having the money now. The longer it takes for FSD to be rolled out the greater the pricing differential needs to be.

Of course, most people are foolish at least some of the time. They may have misjudged both points above, or not even considered it rationally. Such is life.

Now, although I didn't buy FSD, I'd be much happier with a driver assist system than a fully autonomous system. Not least because no one developing autonomy seems to really believe how difficult it is. GM Cruise comes closest, but it is obvious from their own statements that it is superficial -- they don't really believe how difficult it is. This is classic "the last 20% is 80% of the effort" except that it is much more likely to be "the last 1% is 99% of the effort."

And, finally, there is the "small" matter of regulatory approval. In the US expect a lot of feet dragging on Tesla in favor of Waymo and Cruise. There's a lot of money riding on who gets there first and it is so easy to delay approval, or even approve Waymo or Cruise while denying Tesla by cherry picking differences and insisting those are "deal breakers."

A classic economist would not buy FSD today given the numerous delays in the advent of Tesla true self driving. However because Tesla has shifted features that were originally included as part of Enhanced Auto Pilot to Full Self Driving purchases only, buying FSD is not foolish but guarantees you will get the best of Tesla's autopilot software capabilities.

 

[Eno Deb]

A Level 3 vehicle is capable of taking full control and operating during select parts of a journey when certain operating conditions are met. For example, a vehicle that is capable of managing itself on a freeway journey, excluding on- and off-ramps and city driving, might be considered Level 3 automated

Level 3 means that the car can drive without human supervision under some limited conditions. As long as Tesla requires the driver to be attentive and supervise the car at all times, it is not Level 3. This isn't only about the the situations that the car can "manage", but also about liability (which is a measure of the confidence that the carmaker has in the reliability of the system).