Navigate on Autopilot is Useless (2018.42.3)

[dfwatt]

I think the problem with all this is like all anecdotal data and reports it's literally all over the place. I can find YouTube channels dedicated to recording of full self-driving from locations in California to another location that involves 30 miles of traffic without incident. Then I hear horror stories. What we lack is systematic data and only Tesla has access to that.

I believe my early version of autopilot which includes enhanced autopilot but not full self driving on a hardware 2.5 is good but not great. I would never take a long trip without it. It's a godsend on long drives but I don't see it as Uber reliable particularly around Phantom braking which has improved significantly but not disappeared. Absent someone getting Tesla's data perhaps somebody can survey a large group of FSD users?

What I don't understand is somebody seeing FSD as a Make It or Break a deal for whether or not to own a Tesla.

 

[arghx7]

I'm on 11.3.3 FSD BETA, and NoA tries to send me into cones and misses exits, or makes panicked last minute lane changes.

 

[AWDtsla]

Same, 11.3.x misses offramps often. Lane changes towards exits are sometimes improved, sometimes not. I'm beginning to suspect no one on the Autopilot team is a native driver (i.e. grew up in the US, learned to drive on things like go-carts even before license age, etc). FSD's driving probably reflects how they would drive (badly). The dry steering, please just make it stop...

 

[DrDabbles]

What did you get instead? Does it have something like FSD?

Rivian. Thankfully Driver+ is usable and they aren't trying to do anything nonsensical like FSD.

I think the problem with all this is like all anecdotal data and reports it's literally all over the place.

That's actually the crux of the problem. If this was a solved problem in 2016, why are we still having this conversation? This thread is from a 2018 release and we're still seeing posts of NoAP doing absurd things. Remember that there was going to be a coast-to-coast demo in 2017, too.

I absolutely forgive slipped timelines, and as someone that has done everything from hardware design to software development, I can absolutely appreciate an amateur underestimating the challenge that solving a problem presents. But I can not abide a company continually claiming a vehicle that can't handle basic lane changes consistently is going to completely drive itself within the year for seven years straight.

 

[drtimhill]

Same, 11.3.x misses offramps often. Lane changes towards exits are sometimes improved, sometimes not. I'm beginning to suspect no one on the Autopilot team is a native driver (i.e. grew up in the US, learned to drive on things like go-carts even before license age, etc). FSD's driving probably reflects how they would drive (badly). The dry steering, please just make it stop...

That's why it's still beta.

it's worth remembering that the AP team has replaced the entire legacy TACC/NoA stack with a new much more capable one, and this is the first release, so it's hardly surprising it's still somewhat shaky. Frankly I'm surprised it's as good as it is. The old TACC/NoA stack was a dead-end from a development standpoint, while the new stack is a much more solid foundation for building on and, even in its infancy, is pretty much on a par with the old stack. Also, by removing the old stack they free up compute resources for the new stack, which gives them more room to improve the existing networks and add new ones. That's really why V11 is big, not for the immediate improvements, but for the groundwork for future stuff.

 

[verygreen]

's worth remembering that the AP team has replaced the entire legacy TACC/NoA stack with a new much more capable one, and this is the first release, so it's hardly surprising it's still somewhat shaky

and they did it for like 4th time in the past 5 or so years and will keep doing it going forward and as such you will always have a chance to use this excuse.
The problem is: it does not really help anybody that just want the functionality they (believe) they paid for.

 

[Twiglett]

and they did it for like 4th time in the past 5 or so years and will keep doing it going forward and as such you will always have a chance to use this excuse.
The problem is: it does not really help anybody that just want the functionality they (believe) they paid for.

It's this that Elon and Tesla have a blind spot for. Instead of delivering a finished feature and stop calling it beta, they keep adding more & more incomplete features to the scope without finishing anything that was previously delivered.

 

[drtimhill]

and they did it for like 4th time in the past 5 or so years and will keep doing it going forward and as such you will always have a chance to use this excuse.
The problem is: it does not really help anybody that just want the functionality they (believe) they paid for.

Didn't see me giving any excuse, just describing the actual changes and the rationale they applied. A lot of people paid for FSD, and Tesla are pushing hard to deliver that .. or do you know some magic short cut to deliver this functionality that NO-ONE has ever attempted before?

Sure, it's late based on guess-work from everyone from me to Elon. Welcome to the bleeding edge.

 

[Electroman]

Didn't see me giving any excuse, just describing the actual changes and the rationale they applied. A lot of people paid for FSD, and Tesla are pushing hard to deliver that .. or do you know some magic short cut to deliver this functionality that NO-ONE has ever attempted before?

Sure, it's late based on guess-work from everyone from me to Elon. Welcome to the bleeding edge.

Thanks for injecting some reality into this whine fest.

As they keep progressing they hit a wall and then pivot and take different approach based on the experience they have gained.

Like a Lewis&Clark expedition, they know the destination is reachable, just the path has not been laid out or travelled by anyone else.

 

[Dewg]

Thanks for injecting some reality into this whine fest.

As they keep progressing they hit a wall and then pivot and take different approach based on the experience they have gained.

Like a Lewis&Clark expedition, they know the destination is reachable, just the path has not been laid out or travelled by anyone else.

It's impressive and frustrating at the same time. Many companies will pick a trajectory and follow it. If they find issues in the middle of their development, they patch and continue as best they can. Tesla seems more trial and error. They pick a path, and if that doesn't seem to work as well they pivot (as you say) and try a different approach. The downside is that each time they do this there are regressions as they have to train the new systems.

I think they also are amazing at optimizing for the CPU constraints in HW3. They are squeezing more and more capability into the 144 TOPS of HW3. We'll see what HW4 will be able to run pretty soon I'm sure.

 

[clydeiii]

and they did it for like 4th time in the past 5 or so years and will keep doing it going forward and as such you will always have a chance to use this excuse.
The problem is: it does not really help anybody that just want the functionality they (believe) they paid for.

Gotta go downhill to ultimately get off the local maxima and find the global maxima…

 

[verygreen]

Didn't see me giving any excuse

"they just replaced the codebase, that's why it's flaky" is the excuse you gave.

Gotta go downhill to ultimately get off the local maxima and find the global maxima…

gotta use fancy buzzwords to find excuses for one's favorite toy.

 

[clydeiii]

"they just replaced the codebase, that's why it's flaky" is the excuse you gave.


gotta use fancy buzzwords to find excuses for one's favorite toy.

Nah just standard computer science/maths terms.

Btw that is what we’re talking about here, computer science, not toys.

 

[verygreen]

Nah just standard computer science/maths terms.

Btw that is what we’re talking about here, computer science, not toys.

so what are you measuring (and how) that was reaching local maxima in v10 and got downhill with v11?

 

[drtimhill]

"they just replaced the codebase, that's why it's flaky" is the excuse you gave.

No, I gave an explanation. An explanation details why something is in a certain state, an excuse is an attempt to justify it being in that state.

And I said "shaky" not "flaky", don't play the mis-quote game.

 

[clydeiii]

so what are you measuring (and how) that was reaching local maxima in v10 and got downhill with v11?

I’m not sure. Many different variables you can measure. Ideally you’re moving in a direction that increases as many as possible

 

[drtimhill]

so what are you measuring (and how) that was reaching local maxima in v10 and got downhill with v11?

What are you measuring to argue that there has been little/no progress?

 

[verygreen]

What are you measuring to argue that there has been little/no progress?

number of autonomous Tesla cars on the road since 2016.

 

[verygreen]

I’m not sure. Many different variables you can measure. Ideally you’re moving in a direction that increases as many as possible

That does not answer my question, just evades it. It's computer science, and you clearly see the departure from local maxima, so you should be able to explain what and how you are measuring to see this, pretty simple, right? Otherwise it's just buzzwords that you repeat pecause they sound cool or whatever.

 

[drtimhill]

number of autonomous Tesla cars on the road since 2016.

Ah right, of course. All those silly cars running FSD beta are in no way an advance toward such a goal.