AP1 scare

[cdsmith]

Had what I assume was an AP1 malfunction this morning. 2015 MS 90D - software 2018.50.6

Was in stop/go, AM commuting traffic, 4-lane highway, near Philadelphia. AP1 was engaged, I had hand on the wheel, was paying attention. Cruise speed set to 60, but not going anywhere near that pace - had slowed down from that speed automatically as traffic increased, in the center lane. Was approx 2 car lengths behind car in front when he moved to right lane. That opened up a good space in front - maybe 8 - 10 car lengths? My car then accelerated quickly to close the gap, but did *not* slow down as it approached the car in front - small white SUV. It was closing in hard - as in that feeling you get as the passenger when you think the driver doesn't see the stopped car in front of you - Holy sh*t!! I had to brake *hard*, ABS kicked in, screech, stuff on seats now in footwell, etc. stopped maybe 5 feet behind him. Not a reassuring feeling.

I have been using AP for a long time - never had an issue like this. I can't think of anything that was unusual about the situation. Even if it was, that was a textbook case of where AP is supposed to be used as I understand it - multilane, restricted highway.

I iove the car, big fan of tesla and AP, etc. Not hating or looking to throw shade. Just a little freaked, and hesitant to rely on AP now.

Wondering if anyone else has ever experienced something like this?
Should I (somehow) let tesla know so they can investigate? Pull logs?

 

[Mark_T]

Was the small white SUV showing on the radar at all?

If it was stationary the whole time and your car had not previously seen it move...

... or was traffic still in motion at the time... ?

 

[RDoc]

I've had that kind of thing happen as well. Now I always set the speed way down when I'm in stop and go traffic just for that reason.

 

[kavyboy]

That's kind of normal. When the car in front moves to another lane, AP1 will see that as the lane opening up and try to accelerate to cruising speed. You hope it sees the car in front. There's no guarantee that it will be able to see it in time, especially if that car hasn't moved in a while. White vehicles also anecdotally give it a hard time.
When the car in front changing lanes, I either disengage AP or hover my foot over the brake.

 

[cdsmith]

Don't recall if the SUV in front was moving or stopped. If it was moving, it was slow. It was in front of me the entire time after the lane was opened up.

 

[Eevee]

I've had that kind of thing happen as well. Now I always set the speed way down when I'm in stop and go traffic just for that reason.

Yes, had this experience of the OP. So I do the same thing and set my TACC to lower speeds for the stop-and-go traffic.

I think it's clearly a programming error by Tesla. And there are many of them. The car is a great electric car. But self-driving? nope

 

[appleguru]

A stopped car and an overhead bridge/sign have very similar radar signatures I’m told. So if your car never saw it move and it was stopped under a road sign/other “whitelisted” object, it could have easily filtered it out...

 

[vickh]

That's kind of normal. When the car in front moves to another lane, AP1 will see that as the lane opening up and try to accelerate to cruising speed. You hope it sees the car in front. There's no guarantee that it will be able to see it in time, especially if that car hasn't moved in a while. White vehicles also anecdotally give it a hard time.
When the car in front changing lanes, I either disengage AP or hover my foot over the brake.

So any quick swerves by someone in infront of you could cause problems. Does AP2.0 fix that?

 

[vickh]

Yes, had this experience of the OP. So I do the same thing and set my TACC to lower speeds for the stop-and-go traffic.

I think it's clearly a programming error by Tesla. And there are many of them. The car is a great electric car. But self-driving? nope

Elon says end of this year for FSD.

 

[kavyboy]

So any quick swerves by someone in infront of you could cause problems. Does AP2.0 fix that?

I think it has to see the car move out of the lane pretty much entirely. Just moving across the lines a bit is OK.

 

[shred86]

If I'm not mistaken AP primarily uses the radar to detect closure on objects. The issue is stationary objects are problematic for radars as one object might have the same radar cross section as another (i.e. a vehicle, metal can, road signs, metal plates from road construction, etc.), so stationary vehicles will often get rejected/ignored. I'd definitely recommend canceling AP in those situations until you stop behind the vehicle in front of you or it begins moving, then re-engage AP.

Even if traffic is moving, I personally can't stand how quickly AP tries to accelerate in that situation you were in. I wish it did a better job of taking into account the multiple slow vehicles in the adjacent lane and distance to the vehicle in front of you, to have a slower acceleration profile.

 

[vickh]

If I'm not mistaken AP primarily uses the radar to detect closure on objects. The issue is stationary objects are problematic for radars as one object might have the same radar cross section as another (i.e. a vehicle, metal can, road signs, metal plates from road construction, etc.), so stationary vehicles will often get rejected/ignored. I'd definitely recommend canceling AP in those situations until you stop behind the vehicle in front of you or it begins moving, then re-engage AP.

Even if traffic is moving, I personally can't stand how quickly AP tries to accelerate in that situation you were in. I wish it did a better job of taking into account the multiple slow vehicles in the adjacent lane and distance to the vehicle in front of you, to have a slower acceleration profile.

does AP 2.0+ fix that?

 

[shred86]

does AP 2.0+ fix that?

No idea. The issue I mentioned with radars in general isn't something you can necessarily "fix", it's just the way radar works; there's a lot of ambiguities. Tesla could obviously modify their algorithms but the alternative would be your AP randomly slamming the brakes for false positives, which obviously wouldn't be good. I have no idea what is possible with the cameras. I'm sure Tesla could do some magic with correlating objects identified with the camera and radar to help, but who knows.

 

[kgroschi]

Had what I assume was an AP1 malfunction this morning. 2015 MS 90D - software 2018.50.6

Was in stop/go, AM commuting traffic, 4-lane highway, near Philadelphia. AP1 was engaged, I had hand on the wheel, was paying attention. Cruise speed set to 60, but not going anywhere near that pace - had slowed down from that speed automatically as traffic increased, in the center lane. Was approx 2 car lengths behind car in front when he moved to right lane. That opened up a good space in front - maybe 8 - 10 car lengths? My car then accelerated quickly to close the gap, but did *not* slow down as it approached the car in front - small white SUV. It was closing in hard - as in that feeling you get as the passenger when you think the driver doesn't see the stopped car in front of you - Holy sh*t!! I had to brake *hard*, ABS kicked in, screech, stuff on seats now in footwell, etc. stopped maybe 5 feet behind him. Not a reassuring feeling.

I have been using AP for a long time - never had an issue like this. I can't think of anything that was unusual about the situation. Even if it was, that was a textbook case of where AP is supposed to be used as I understand it - multilane, restricted highway.

I iove the car, big fan of tesla and AP, etc. Not hating or looking to throw shade. Just a little freaked, and hesitant to rely on AP now.

Wondering if anyone else has ever experienced something like this?
Should I (somehow) let tesla know so they can investigate? Pull logs?

Well it's not perfect but that's why you gotta pay attention. This sort of thing happens with pretty much any active cruise control. I won't even let the car accelerate hard when the car in front changes lanes. I will wait for a second and then disengage immediately or adjust the max speed because you gotta be always one step ahead. Glad you didn't risk it all and get into another AP crash. That would be bad for all of us

 

[maximizese]

That is flat out scary. I'm glad you posted this to warn and confirm with others. It is for experiences like this that I do not trust AP. It works, but it only works within the parameters in which it was programmed. Any blindspots that the programmers failed to recognize will translate into a flawed system and that doesn't sit well with me. I'm glad you guys are figuring out the bugs and hopefully there will be updates to fix these problems. Just imagine all the folks who are using the feature without the knowledge of how it works and the instance in which it does not. I'm glad the OP is safe. I think we've all had those moments when you feel your heart sink, time slows down, and you think the worst is about to happen.

 

[BridgeMojo]

Sounds like a variation of this problem. It's great to be aware of it:

Why Tesla's Autopilot Can't See a Stopped Firetruck

 

[vickh]

Sounds like a variation of this problem. It's great to be aware of it:

Why Tesla's Autopilot Can't See a Stopped Firetruck

can AP2.0 "see" further ahead of 1 car?

 

[whitex]

does AP 2.0+ fix that?

Not for AP1 cars. It's supposed to be fixed for AP1 cars once it's no longer beta - which in all likelihood will not happen before all AP1 cars turn into dust.

 

[whitex]

To all using Auto Pilot, AP1, AP2 or AP2.5 - this is a called a beta feature for a reason, it is experimental and you are the voluntary test pilot. So treat it like a test pilot would treat an experimental aircraft - be on your toes at all times and expect the unexpected (though in this case it's kind of expected by people who use it often enough).

 

[aaron0k]

To all using Auto Pilot, AP1, AP2 or AP2.5 - this is a called a beta feature for a reason

Usually a beta is followed by a final release... since they abandoned further development on AP1, does that mean it will always be in "beta"? Can I get my money back since it's clear it will never be 'finished'?

Crazy things happen even in final release software... however, putting it in a production car (that can kill you), charging $5,000 for it and perpetually calling it "beta" is... bullshit.

Tesla uses the "beta" label to limit their liability.