Navigate on Autopilot is Useless (2018.42.3)

[scottf200]

We'd need someone that has disassembled the firmware to comment. But basically AP as far as I can tell is the combination of lane keep assist, and TACC. They all behave the same to me, and I frequently use TACC rather than AP in areas where AP just makes terrible decisions. TACC also fails to operate when the AP computer crashes, which would again seem to indicate that it's using the same inputs and is almost certainly the same system.

Re: underline text --- Are you conflating AP with NoA? AP is just Autosteer+TACC so centering and staying behind. No decisions.

TACC - I used this in the suburbs or on roads where the majority of the traffic wants to go > speedlimit+5 (ie. +10
AP - I use this in all suburb driving and basic highways (non-divided)
NoA - I use this on most interstates and roadtrips. AP + decision making (ie. off ramp auto-exits are handy for day dreaming or if in conversations with passengers)

 

[DrDabbles]

Re: underline text --- Are you conflating AP with NoA? AP is just Autosteer+TACC so centering and staying behind. No decisions.

TACC - I used this in the suburbs or on roads where the majority of the traffic wants to go > speedlimit+5 (ie. +10
AP - I use this in all suburb driving and basic highways (non-divided)
NoA - I use this on most interstates and roadtrips. AP + decision making (ie. off ramp auto-exits are handy for day dreaming or if in conversations with passengers)

No, I mean AP. Like it sees a lane split and instead of continuing forward it does something random. When two lanes merge, it forces itself to center in the new double wide lane pretty much no matter what. When a vehicle is merging onto the highway and would (and legally should) end up behind me, AP jacks on the brakes to let them in first regardless of the speed differential and any traffic following me. And of course, predictable phantom braking. AP on its own makes poor decisions.

If you'd like me to tell you my thoughts on NoA (much worse behavior than AP), I'd be glad to. I'm not new to this, so I'm pretty up to snuff on which system is which, and what it does or doesn't do.

 

[eli_]

No, I mean AP. Like it sees a lane split and instead of continuing forward it does something random. When two lanes merge, it forces itself to center in the new double wide lane pretty much no matter what. When a vehicle is merging onto the highway and would (and legally should) end up behind me, AP jacks on the brakes to let them in first regardless of the speed differential and any traffic following me. And of course, predictable phantom braking. AP on its own makes poor decisions.

If you'd like me to tell you my thoughts on NoA (much worse behavior than AP), I'd be glad to. I'm not new to this, so I'm pretty up to snuff on which system is which, and what it does or doesn't do.

I don't believe that AP even perceives forks or merges, like the vision system doesn't detect them. Could be wrong but it seems to rely on map data to know when a car in an adjacent lane will be merging into yours. Obviously map quality varies wildly from one place to another.

 

[clydeiii]

Can you point out the timestamps in my video where the car would have annoyed you?

 

[GlmnAlyAirCar]

Can you point out the timestamps in my video where the car would have annoyed you?

Wow do you expect anyone to watch a 31 minute video?

 

[clydeiii]

I don't really but people asked me to record it in order to prove that I wasn't delusional.

 

[arghx7]

2019.40.2.1 solidly moves NoA out of the "useless" category. It still requires some interventions once in a while (car doesn't "see" merging cars coming from the side that well) but the improved lane changes are a real benefit.

 

[mtndrew1]

2019.40.2.1 solidly moves NoA out of the "useless" category. It still requires some interventions once in a while (car doesn't "see" merging cars coming from the side that well) but the improved lane changes are a real benefit.

Same opinion here. The last few months of development, and particularly this firmware, have moved NoA solidly away from useless in my use case.

I don’t know that it’s moved all the way over to the useful category quite yet but it’s very close. It’s like autopilot purgatory.

 

[rnortman]

2019.40.2.1 solidly moves NoA out of the "useless" category. It still requires some interventions once in a while (car doesn't "see" merging cars coming from the side that well) but the improved lane changes are a real benefit.

Lane changes have not been the problem for me. Knowing what lane to be in when has always been the problem. Has this improved?

Well, lane changes are also a problem, because around here there are interchanges where you need to cut across 3-4 lanes of traffic in 0.5 miles in order to make your exit, and the auto lane change just doesn't have a chance. I would be shocked if this were not still a problem.

 

[arghx7]

it will effectively pass on the right sometimes, which isn't acceptable in some jurisdictions. Since it changes lanes better it handles those 3-4 lane changes better. The other thing is that people's perceptions of the "correct" lane to be in depends on their own driving style. Some people really don't mind squeezing through 3 lanes of traffic pretty close to an exit, and others are more likely to get over earlier.

Yesterday I let it try to merge into a stop and go interchange, but it tended to put on the signal and just sit there and block traffic, waiting for an opening (I took over). This is hard for an automation system to handle because humans can see far ahead and will use hand signals with other drivers to get in, or even jump a solid line if they see an opening.

 

[clydeiii]

I'm enjoying the 40.x "aggressive" lane changes so far but it's been snowy or rainy in DC the last few days, which has caused NoA to turn on and off over and over, not letting me see the true capabilities of the new feature so far. Maybe this evening's commute will give it a chance to shine.

 

[eli_]

Given how frequently adjacent cars can go missing or bounce between lanes on the ADAS visualization, the quick lane changes make me uneasy.

 

[scottf200]

Given how frequently adjacent cars can go missing or bounce between lanes on the ADAS visualization, the quick lane changes make me uneasy.

Seems like a silly comment. How many 10s of 1000s of 'quick lane changes' do you think all of the 40.2.1 cars (beta and in the past 3-4 days) have made? My display of adjacent cars is quite solid but I don't think it is reflecting exactly what the sensors see. It is being made UI friendly and avoiding overloading users.

2019.40.2.1 solidly moves NoA out of the "useless" category. It still requires some interventions once in a while (car doesn't "see" merging cars coming from the side that well) but the improved lane changes are a real benefit.

Same opinion here. The last few months of development, and particularly this firmware, have moved NoA solidly away from useless in my use case.
I don’t know that it’s moved all the way over to the useful category quite yet but it’s very close. It’s like autopilot purgatory.

I'm enjoying the 40.x "aggressive" lane changes so far but it's been snowy or rainy in DC the last few days, which has caused NoA to turn on and off over and over, not letting me see the true capabilities of the new feature so far. Maybe this evening's commute will give it a chance to shine.

I just did a 250 mile trip today on a mixture of rural roads, major highways, and 25K 'city' driving all while using AP and NoA when applicable. It worked very well.

I did, however, notice that the 'adaptive cruise control' aspect of following a car seems a little herky-jerky (or speed up, down, up, down) compared to the car I'm following.
ScanMyTesla data on my '17 X 100D. Obviously speed changed some but the purple HP line gives you and idea that it wasn't smooth and you could feel it.

https://i.imgur.com/hPvYppI.jpg

https://i.imgur.com/2a8WbZC.jpg

 

[mtndrew1]

Seems like a silly comment. How many 10s of 1000s of 'quick lane changes' do you think all of the 40.2.1 cars (beta and in the past 3-4 days) have made? My display of adjacent cars is quite solid but I don't think it is reflecting exactly what the sensors see. It is being made UI friendly and avoiding overloading users.




I just did a 250 mile trip today on a mixture of rural roads, major highways, and 25K 'city' driving all while using AP and NoA when applicable. It worked very well.

I did, however, notice that the 'adaptive cruise control' aspect of following a car seems a little herky-jerky (or speed up, down, up, down) compared to the car I'm following.
ScanMyTesla data on my '17 X 100D. Obviously speed changed some but the purple HP line gives you and idea that it wasn't smooth and you could feel it.

https://i.imgur.com/hPvYppI.jpg

https://i.imgur.com/2a8WbZC.jpg

Thank you for this post. This aggressive accel/decel is making me sick to my stomach. This is the first time I’ve voluntarily driven my car manually in heavy traffic; I can’t handle the lurching on this firmware.

 

[scottf200]

Thank you for this post. This aggressive accel/decel is making me sick to my stomach. This is the first time I’ve voluntarily driven my car manually in heavy traffic; I can’t handle the lurching on this firmware.

You are most welcome. I just felt it myself today and I normally am not sensitive to it but it was noticeable. I *really* need to do a test with a friend that I can follow while they are on a steady ACC speed. I was just trying it for a short while on the highway tonight.

 

[timewasted]

2019.40.2.1 AP2.0: it’s trying to be useful, but still useless. My speed limit was set to 75, and there was a big truck a decent distance in front of me going slower, so it offered to move me to the faster lane to the left. Finally, it made a seemingly intelligent decision for the first time since this feature rolled out! Unfortunately, even though this left lane was open and clearly going faster than the previous lane, not even 5 seconds later it put me back behind the still slow moving truck because it thought that was the faster lane.

So it’s at least progress, but only in the sense of it went from s*itting the bed 100% of the time to s*itting the bed about 95% of the time. Yay?

 

[DrDabbles]

Lane change decisions are still pretty bad. NoAP still wants to put you into an exit lane almost 2 miles before your exit, which is a completely insane choice in my region where exits may be less than half a mile apart. This means I enter a lane that's basically stop-and-go for 2 miles while the rest of the highway flies by.

And of course, the change to one lane only to immediately change back to the original lane that you mentioned. I turned lane change confirmation off with 40.2.x to see how it behaved, and I've turned confirmation back on again. It's not ready for prime time.

 

[diplomat33]

Lane change decisions are still pretty bad. NoAP still wants to put you into an exit lane almost 2 miles before your exit, which is a completely insane choice in my region where exits may be less than half a mile apart. This means I enter a lane that's basically stop-and-go for 2 miles while the rest of the highway flies by.

My experience is the opposite. It often changes lanes too late for my comfort (less than 1.5 mile before my exit) and it makes me very nervous that I am going to miss exit when traffic is congested.

 

[DrDabbles]

My experience is the opposite. It often changes lanes too late for my comfort (less than 1.5 mile before my exit) and it makes me very nervous that I am going to miss exit when traffic is congested.

And this is why dumb robots aren't the solution, and is precisely the point MobilEye has been yelling about for basically two decades.

 

[clydeiii]

And this is why dumb robots aren't the solution, and is precisely the point MobilEye has been yelling about for basically two decades.

Why do you assume Tesla is building a "dumb robot" solution here?