Navigate on Autopilot is Useless (2018.42.3)

[Jeff Hudson]

Aside: yes, I have pressed the accelerator pedal and gone 75ish in this equivalent example

I have done this same thing as well while just using Autopilot. Tesla owners with Autopilot and/or NoA have been learning a new set of 'driving responses' possibly not expected to be learned when we originally bought our Tesla's. Pressing the accelerator pedal to facilitate a lane change is a learned response just the same as cancelling Autopilot or NoA when appropriate.

 

[AwlBidnz]

It’s a learning process for both sides of the MCU ;0)

 

[verygreen]

why can't it just smoothly change over to the left without slowing down?

It really wants a bigger speed difference before initiating the change. I think if you set it to madmax it would be more proactive. (I miss the days where you can just set the actual numerical speed offset)

When I approach something with 15-20mph speed difference, it changes waaay ahead of time even in v9 and so there's no slowdown.

 

[AwlBidnz]

It really wants a bigger speed difference before initiating the change. I think if you set it to madmax it would be more proactive. (I miss the days where you can just set the actual numerical speed offset)

When I approach something with 15-20mph speed difference, it changes waaay ahead of time even in v9 and so there's no slowdown.

I tried mad max as well as increasing the difference to the car in front, which I was thinking that was working. It didn’t. v10 seems worse?

 

[wk057]

I tried mad max as well as increasing the difference to the car in front, which I was thinking that was working. It didn’t. v10 seems worse?

Same. I've tried every combination of settings. No matter what, the car will always slow down before initiating a lane change when overtaking another car. Doesn't matter the speed delta, the autopilot "aggressiveness" setting (1-7), the lane change aggressiveness setting (madmax, etc), whatever. It will see the car ahead. Approach it, slow, and only after slowing at least a little will it make a lane change. I've seen some videos where occasionally it will just barely start the lane change before slowing, but I've not experienced this in all of my use of it.

I mean really, this was like the first thing I implemented in my hacked AP1 auto lane change stuff: smooth and fluid lane changes without slowing. It's not complicated, and the AP2 setup even seems to see the car-to-be-overtaken much sooner than my AP1 setup, yet still fails to do this relatively simple thing.

As soon as the car comes into view of the AP system, it knows your speed delta.

 

[timewasted]

Going from 2019.8.4 to 2019.8.5 resulted in exactly zero improvement in navigate on autopilot. My car shows no real desire to make an intelligent decision, and it’s certainly in no rush to make it to its destination. I turned off the need for confirmations, and each time I’ve left work the need for confirmations was turned back on.

The stop and go traffic handling seems to be smoother in this version (did we finally get our “silky smooth” update?!), but the marquee feature is still proving useless to me.

In the six months that have passed since posting this, I’ve seen a massive change... in the version number. Every complaint that I’ve posted in this thread still holds true today.

 

[verygreen]

Same. I've tried every combination of settings. No matter what, the car will always slow down before initiating a lane change when overtaking another car.

I know you saw this video, but others might benefit from seeing it too.

VID_20190825_105149 - Streamable

That said NoA is still a mostly useless party trick in my view.

 

[Electroman]

I know you saw this video, but others might benefit from seeing it too.

VID_20190825_105149 - Streamable

That said NoA is still a mostly useless party trick in my view.

Driving at 90 on AP.. thats reckless.

 

[clydeiii]

In the six months that have passed since posting this, I’ve seen a massive change... in the version number. Every complaint that I’ve posted in this thread still holds true today.

In the 6 months since the original post, Nav-on-AP has only gotten better and better with each release ... for me.

 

[Electroman]

In the six months that have passed since posting this, I’ve seen a massive change... in the version number. Every complaint that I’ve posted in this thread still holds true today.

I am guessing your time is wasted checking out NoAP ?

 

[S4WRXTTCS]

It's getting to the point where TMC members should group together to test NoA over various 100+ mile stretches, and simply record pass/fail data.

The most important part is to remove the interpretation element, and some of the regional elements.

So it's crystal clear what's being tested, and for the most part it's tested against the previous version.

Like a good example would be whether the car tries to re-center itself in the right lane during merge points. This happens most of the time when I try it in WA state.

To this day I consider NoA a failure because of that behavior.

I've also experienced the slowing by 1-2 mph AFTER passing a vehicle and getting in that lane, and it happens regardless of how far away the the next vehicle ahead of me is.

I think if we members did a better job collection one dataset of pass/fail issues that we'd have better luck getting those things addressed in future updates.

 

[S4WRXTTCS]

That said NoA is still a mostly useless party trick in my view.

Who's going to have a party in the middle of nowhere?

The only place I find NoA useful is on long stretches of road with little traffic, but enough traffic that you frequently have to pass slow people to the right.

It's especially nice on 3 lane roads where you can spend most of the time in the middle lane passing trucks to the right, and getting passed by crazy people doing 90 in the left lane.

On 2 lane roads it depends on the frequency of the merging lanes. If they're frequent, and of the standard type then AP itself is a pain.

I do find AP/NoA more enjoyable in Oregon because they paint dashed lines to the right on the merging points so AP doesn't re-center itself.

 

[DrDabbles]

My state mixes dashed lines in merges and no dashed lines. So it's a real fun surprise to passengers when we suddenly jerk four feet to the right and then drift back left.

 

[verygreen]

The only place I find NoA useful is on long stretches of road with little traffic, but enough traffic that you frequently have to pass slow people to the right.

would it inconvenience you too much to manually request a lane change in those conditions?

 

[Electroman]

would it inconvenience you too much to manually request a lane change in those conditions?

Exactly !

Sometimes the solution is simple and staring right at us. We just have to some common sense to realize and use that.

 

[AwlBidnz]

You will eventually accept the training your car is providing & start turning off AS before those missing paint, merging lanes. Now if they could jump up joe mode to silence the bong bong, we could drive without alerting our passengers ;0-

 

[S4WRXTTCS]

would it inconvenience you too much to manually request a lane change in those conditions?

One of the selling points that attracted me to enhanced autopilot was automatic lane changes that I didn't have to initiate. So to answer this question I'd say yes it would because that defeats the entire purpose of getting this package.

There is also certain elegance/beauty of it during those moments, and there is enjoyment in that.

 

[S4WRXTTCS]

Exactly !

Sometimes the solution is simple and staring right at us. We just have to some common sense to realize and use that.

It's hilarious that what is essentially a negative post about NoA wasn't negative enough.

verygreen stated that NoA was mostly a useless party trick in his view.

So I made a joke on that because in my view it's only useful out on the middle of nowhere. Who's going to have a party in the middle of nowhere? It's not like you can say "Hey, let me show you this cool feature but we'll have to go on a 50 mile drive to show you".

That was largely the extent of my response. I hardly expected a serious conversation to develop from it.

I fail to understand your response at all. Why would I be looking for a solution for something that actually worked?

If it did the job why would I put 10% more effort into pushing a button? Why would you be questioning another mans laziness? I take my laziness seriously. It's 50% of the reason I got an EV in the first place.

Common sense is not to use a feature where you put others in danger. How is that applicable to someone who simply said where they found NoA useful?

The sad reality is that NoA hasn't really delivered on it's promises.

It doesn't fix AP as the car still tries to recenter itself on what are clearly merges/exits for things NoA should be telling it to ignore.
It doesn't have the situational awareness to be all that useful in crowded areas
It suffers from the weaknesses in other systems. For example if the car is doing an automatic lane change into the middle lane from the left it can easily get confused about what lane the car in the far lane is in. Sometimes it puts it into the middle lane, and then the lane change gets canceled. The new visualizations show it really well. And, it actually happens fairly frequently. The entire identify/visualization neural network is broken with V10. Things are constantly going in and out. The whole HW2/HW2.5 needs to go into the recycle bin or re-purpose bin.

But, all the above doesn't remove the fact that there are traffic situations where it does work well, and it can be a hundred miles or more of this with 50+ automatic lane changes done automatically with no drama.

I enjoy NoA in these low traffic situations. In those parts I can actually enjoy what I paid for. I wish the people in front of me in the left lane had that feature so their car would automatically move them out of the passing lane.

 

[eli_]

NoA just exposes how bad the state of highway map data is. Lots of navigation apps have lane guidance where they highlight what lane you should be in, and most people ignore it and there's no real consequence when it's wrong as long as the general directions are accurate. But with NoA, the map database is now Truth, and the basis for how it plans all maneuvers. I guess what's really needed is a system that parses highway navigational signage and relying less and less on the map data, which is always going to be wrong in some cases. It'll just take a few years to develop.

 

[verygreen]

So I made a joke on that because in my view it's only useful out on the middle of nowhere. Who's going to have a party in the middle of nowhere?

for the party trick part of it, you don't need to travel far, just punch in any route into the nav that would get you on an interstate highway as part of it.

I don't really know why do they need NoA to be active in order to activate ULC to maintain desired speed, btw, seems like a strange requirement.