Forums

AP didn't stop when car switched lanes

AP didn't stop when car switched lanes

I noticed when in really slow traffic, a car switched lanes into my lane but I noticed my car in AP didn't try to stop when I was LITERALLY about to hit their side bumper.
I didn't get any pings that I was about to hit something.
Is this an issue with AP?

nukequazar | December 2, 2019

Do you have the latest firmware (2019.40.1.1)? Part of this update was more aggressive lane changes, and I think it's dangerous. I have had similar to what you're describing happen to me twice since this update, although it was when my car was making an aggressive auto-lane change with little pre-signal, and it didn't back off when another car changed lanes into the same lane space. Of course I can't be sure if I would have hit the other car but I did put on the brakes because I was getting very close, and it was rude either way.

drahcirvn | December 3, 2019

I do have the latest update.

But I was not switching lanes, a car was switching into my lane and was 50% in my lane and my car did not even try to stop/slow down.

TeslaTap.com | December 3, 2019

What following distance do you have set? I usually set it to 4 or 5, but if you have it set to a lower value, it will not provide much separation and may contribute to the close call. Also AP today requires paying attention and taking over if needed. It is not FSD yet.

Also if someone dives into your lane with almost zero clearance, it's hard for AP to see that with today's software. It's almost as hard for us humans to react to a crazy driver like this.

jimglas | December 3, 2019

I have mine set at 1. Any farther and I find some other drivers passing me to gain one car length.
The AP reacts faster than I can and the radar sees 2 cars ahead.

RJMIII | December 3, 2019

My engineering opinion is this is the opposite behavior of phantom braking. The situational awareness display shows that the car only has so much certainty about where other cars are around it (i.e. dancing cars). Whenever the NeuralNet identifies a car and its location/velocity it can only be accurate to within some limit. So, sometimes you get braking when you shouldn’t and no braking when you should. Given that phantom braking occurs more than no braking, the system is appropriately set to be conservative.

It will get better with bigger computers (HW3) and better NeuralNets trained on more data. In the meantime, pay attention and believe the Tesla when they call it a driver assist system.