Autonomous Car Progress

[emmz0r]

Feature complete means that it's on the beta stage at the very least.

 

[diplomat33]

Feature complete does not mean primary features are coded. It means all features are coded. They don’t have to be reliable though. Level 5 no geofence has a very long list of requirements...

Nope. Feature complete refers to planned or primary features not all features. See definition below:

"A feature complete version of a piece of software has all of its planned or primary features implemented but is not yet final due to bugs, performance or stability issues.[1] This occurs at the end of alpha testing of development.

Usually a feature complete software still has to undergo beta testing and bug fixing, as well as performance or stability enhancing before it can go to release candidate, and finally gold status."
Feature complete - Wikipedia

 

[electronblue]

@diplomat33 I don’t think ”primary” there means what you think it means (vs. ”secondary” features). As you can see, the rest of the text discusses stability and a release candidate status, not missing feature implementation.

 

[diplomat33]

@diplomat33 I don’t think ”primary” there means what you think it means (vs. ”secondary” features). As you can see, the rest of the text discusses stability and a release candidate status, not missing feature implementation.

Even if that is true, what about the term "planned". Feature complete is all the planned features that the dev team has. So in the case of FSD, feature complete would be all the features that Tesla has planned out. It won't be every single feature in the final product.

 

[electronblue]

@diplomat33 Planned just means whatever was planned for that software version. In this case the aim has been specified as Level 5 no geofence to us so if that was true there is a Level 5 corresponding feature list at Tesla and a realistic roadmap of achieving development completion of that list within 2019.

 

[diplomat33]

@diplomat33 Planned just means whatever was planned for that software version. In this case the aim has been specified as Level 5 no geofence to us so if that was true there is a Level 5 corresponding feature list at Tesla and a realistic roadmap of achieving development completion of that list within 2019.

Yes, I agree. What I am trying to communicate to you is that I think Tesla's "L5 feature complete nogeofence" probably has less features in it that what you think. It won't be every single L5 features. It will just the minimum features needed for a viable L5 prototype with driver supervision. The remaining features will go in AFTER "feature complete" as Tesla works to get L5 to driverless status.

 

[AlanSubie4Life]

With cut-in detection like this...they've still got plenty of work to do...

Cut-in detection...almost

Zero response from the car, until the trajectory of the minivan was about to take it over the lane line. And it's not clear in this case that the response to the minivan was due to its trajectory, or actually due to the braking from the work truck in front. Regardless, still no detection of turn signals that I have observed.

As usual, onset of braking/deceleration can be seen by observing the nose dip. Starts at about 3 seconds.

 

[Eno Deb]

With cut-in detection like this...they've still got plenty of work to do...
[...]
Zero response from the car, until the trajectory of the minivan was about to take it over the lane line.

In many other cases it responds to cars in the next lane that are nowhere near to cutting in and phantom brakes. It all averages out.

 

[AlanSubie4Life]

In many other cases it responds to cars in the next lane that are nowhere near to cutting in and phantom brakes. It all averages out.

If only there were a way to tell when someone was going to change lanes...

Personally I haven't had a problem with phantom braking, but my Autopilot use has been fairly limited.

 

[Eno Deb]

If only there were a way to tell when someone was going to change lanes...

Personally I haven't had a problem with phantom braking, but my Autopilot use has been fairly limited.

In my experience it depends a lot on the traffic pattern. In light and stop-and-go traffic I almost never have it, but heavy moving traffic seems to trigger it a lot, and it always happens when I'm coming up to a car in a neighboring lane. I think it's probably their attempts at reading a car's "body language" (as described by Karpathy) that are going wrong.

 

[Daniel in SD]

In many other cases it responds to cars in the next lane that are nowhere near to cutting in and phantom brakes. It all averages out.

Conspiracy theory time. Maybe Tesla actually adds a lot of random and unpredictable behavior to Autopilot so true believers will post amazing videos about how "Autopilot avoided an accident!", "Autopilot responded faster than a human!", "Autopilot predicted a cut-in!", or "Autopilot navigated a left turn through an intersection perfectly!"

 

[electronblue]

Yes, I agree. What I am trying to communicate to you is that I think Tesla's "L5 feature complete nogeofence" probably has less features in it that what you think. It won't be every single L5 features. It will just the minimum features needed for a viable L5 prototype with driver supervision. The remaining features will go in AFTER "feature complete" as Tesla works to get L5 to driverless status.

While I guess I wouldn’t exactly agree with your characterisations on this as you see them on your end, yes I would agree there is some middle ground where one would be Level 5 no geofence feature complete and still have some ”secondary” features not yet done.

What Tesla meant and what their status is, that is unkown in my view.

 

[diplomat33]

Nice video on Twitter (anyone know how to embed twitter links properly here?) via Teslarati of a HW3 car running an EAP version of Autopilot:

Tesletter - normal driving
Tesletter - stopped at a junction

Few observations:

1. Shows oncoming traffic!
2. Shows the new vehicle categories mentioned earlier in this thread
3. Allows the user to change the viewpoint
4. Super smooth and no jittering cars (except cross traffic)

Some nice improvements being tested in V10 Early Access. Still buggy right now, but having the car correctly track incoming traffic and cross traffic at intersections will be a big step towards "feature complete".

 

[diplomat33]

While I guess I wouldn’t exactly agree with your characterisations on this as you see them on your end, yes I would agree there is some middle ground where one would be Level 5 no geofence feature complete and still have some ”secondary” features not yet done.

What Tesla meant and what their status is, that is unkown in my view.

Yes, there is some middle ground. And that's part of the problem.

The fact is that you and I will have our own personal ideas of what should go into "L5 feature complete no geofence" that probably won't agree with Tesla's internal list. And, we are not privy to Tesla's internal "feature complete" to-do list. So how are we going to judge whether Tesla achieves it or not?

That is why I think it is better to look at the 2 FSD features that are "coming later this year" on the order page. That is something very specific that Tesla is promising. We can judge Tesla on whether they deliver those specific features by end of 2019.

 

[electronblue]

Yes, there is some middle ground. And that's part of the problem.

The fact is that you and I will have our own personal ideas of what should go into "L5 feature complete no geofence" that probably won't agree with Tesla's internal list. And, we are not privy to Tesla's internal "feature complete" to-do list. So how are we going to judge whether Tesla achieves it or not?

That is why I think it is better to look at the 2 FSD features that are "coming later this year" on the order page. That is something very specific that Tesla is promising. We can judge Tesla on whether they deliver those specific features by end of 2019.

Limiting ourselves to just their website is not accurate either since so much of Tesla’s messaging comes elsewhere. We must keep looking at the full picture the best we can. I would agree it may be difficult at times as there are unknowns but that is the only way to strive for accuracy.

One tool is seeing how those 2020 robotaxis materialize, when and where — and what leaks may come. That may give us some guidance on when Tesla reached feature completion, for example.

 

[diplomat33]

Limiting ourselves to just their website is not accurate either since so much of Tesla’s messaging comes elsewhere. We must keep looking at the full picture the best we can. I would agree it may be difficult at times as there are unknowns but that is the only way to strive for accuracy.

One tool is seeing how those 2020 robotaxis materialize, when and where — and what leaks may come. That may give us some guidance on when Tesla reached feature completion, for example.

I am not suggesting we limit ourselves to just the website. However, the website does offer a specific metric that we should not ignore.

Ultimately, I want folks to be cautious. We all know Elon's poor track record in terms of predictions. Likewise, it would be easy to take one comment like Elon's "yes" to "L5 feature complete nogeofence", get a certain idea about it means and then get disappointed if it does not happen on schedule like we thought.

I would agree that our best bet is to focus on what Tesla delivers to the public. And the nice thing about being a Tesla owner is that we will experience the FSD features first hand. We won't need to read about the robotaxis or watch youtube videos about them like with Waymo. When Tesla gets to the point where they can do robotaxis, we will see first hand in our cars, how good they are.

 

[electronblue]

I am not suggesting we limit ourselves to just the website. However, the website does offer a specific metric that we should not ignore.

Ultimately, I want folks to be cautious. We all know Elon's poor track record in terms of predictions. Likewise, it would be easy to take one comment like Elon's "yes" to "L5 feature complete nogeofence", get a certain idea about it means and then get disappointed if it does not happen on schedule like we thought.

I would agree that our best bet is to focus on what Tesla delivers to the public. And the nice thing about being a Tesla owner is that we will experience the FSD features first hand. We won't need to read about the robotaxis or watch youtube videos about them like with Waymo. When Tesla gets to the point where they can do robotaxis, we will see first hand in our cars, how good they are.

I would prefer people to be cautious too — and that includes holding Tesla accountable to what they announce. So I am not looking at these things cautiously just to avoid being misled, I also look at them cautiously to notice and note if we are being misled.

An important distinction that I sometimes feel like your point of view tries to avoid.

 

[diplomat33]

I would prefer people to be cautious too — and that includes holding Tesla accountable to what they announce. So I am not looking at these things cautiously just to avoid being misled, I also look at them cautiously to notice and note if we are being misled.

An important distinction that I sometimes feel like your point of view tries to avoid.

The problem is that we don't always have a basis for accurately judging if we are being misled. Being misled implies motive and intent which we can't always know. We can certainly judge Tesla on missed deadlines. That is more quantitative and objective. For example, we can know if Tesla misses their deadline to release a feature like automatic city driving by end of 2019. But getting to the reasons behind the missed deadline is harder. Did Tesla miss the deadline because of an unexpected setback or because they are deliberately misleading us? That is more subjective.

 

[electronblue]

The problem is that we don't always have a basis for accurately judging if we are being misled. Being misled implies motive and intent which we can't always know. We can certainly judge Tesla on missed deadlines. That is more quantitative and objective. For example, we can know if Tesla misses their deadline to release a feature like automatic city driving by end of 2019. But getting to the reasons behind the missed deadline is harder. Did Tesla miss the deadline because of an unexpected setback or because they are deliberately misleading us? That is more subjective.

Of course it is hard and of course there is a subjective element. Doesn’t mean we shouldn’t be on the lookout for it.

Look. There is history and thus reason for suspicion here. ”Elon is bad at predicting” just doesn’t quite cut it anymore for some of us with the history. I would say that is understandable.

 

[diplomat33]

Brand new marketing video from Waymo with no driver at all in the driver's seat. I do wish that Waymo would put out longer videos that show more of what the car actually is doing. These 1 minute marketing videos are heavily edited and sanitized to only show Waymo in the best possible light.