AP Tends to overshoot when changing lanes, bug?

[RDoc]

The other evening we were driving on a clearly lane striped highway, the display was showing the lanes with no problem, light traffic, good lighting. I changed lanes to the right with the signal stalk, which the car started to execute, and after it got about half way done, I let go of the stalk and the car continued to go into the new lane as normal. However, it kept going into the next lane to the right. There wasn't anyone there, so we let it go to see what would happen.

It got about half way into the next lane, with the lane marker about at the middle of the car. The display screen clearly showed all this as well, with the car icon moving across the lane marker about half way, then correcting back, so the car at some level knew what was happening.

There was another car about 2 lengths ahead in the lane the Tesla erroneously tried to go into, shown in gray on the display, so the car sensed it, but wasn't following it. The lane markers on the display were always clear and blue.

This isn't the first time our Tesla has overshot on line changes with another car in the next lane over. I'm wondering if the CPU is overloaded and/or there's a software bug. It's trying to characterize the car it senses ahead and in the next lane and at the same time trying to complete the lane change.

Not knowing anything about AP's internals but having done some real time programming, I wonder if they are using cooperative multitasking, but have a CPU hog in the code that's not doing enough yields. The thread trying to figure out what the deal is with this new target in front and to the right isn't letting the other thread, which is trying to complete the lane change, run enough. The chip has lots of CPU's, but who knows how the code is organized.

 

[DoubleE]

Mine does that kind of. It switches lanes, and then will drive to the outside of the next lane, sometimes driving on the lane marking, before centering itself again.

It shows all of the lane marking in the cluster, so I'm not sure why it overshoots.

Seems like a definite bug to me.

 

[Electroman]

Never had that happen. I use it at least 7 or 8 times every day and all I have seen is hesitation at times to start the lane change, but once it starts it does it flawlessly.

 

[Larry Hutchinson]

Doesn't the manual say that lane change cancels if the turn signal is released before it is finished?

 

[Electroman]

Not really. There is a point of no return where the lane change will complete even if you cancel the indicator. I believe it is when one of your wheels has crossed the line fully.

 

[AB4EJ]

I noticed a tendency to overshoot when lane-changing in AP; what I have started doing is to flip the turn signal stalk back to neutral position as soon as the car is about 1/3 of the way into the next lane, and then it completes the lane change and doesn't overshoot. (Although I must say I am not sure that this action is really the cause of the non-overshoot, it may be that they lane change was going to be fine even without it).

 

[sorka]

I've used auto lane changing 50+ times and it has never done anything but do it exactly like I do. Has never overshot the middle of the target lane for me.

 

[RDoc]

Most of the time the lane change works perfectly for me too. However, what I'm suggesting is that if the system is doing the lane change and discovers another car in the next lane over from the targeted lane while in the middle of that process, it gets overwhelmed.

It's like having trouble walking and chewing gum at the same, in this case, simultaneously completing the lane change and evaluating the newly discovered vehicle.

 

[sorka]

Haven't noticed any issues with other cars either. Have you reported it? Do you have dash cam?

 

[jzzsxm]

Mine does this as well - it acts underdamped, to the point where I'm pretty sure lane changing is driven by basic PID controls. The P and the I are fine, but if somebody at Tesla could fiddle with the D, that'd be great. Thanks.

 

[Electroman]

I know that graph is self explanatory to a lot folks here, but a little explanation for the rest of us would help. Like perhaps, I am sure 'c' is the speed of light, right?

 

[mblakele]

I'm not convinced that auto lane change has any literal PID controllers in its implementation, but PID controller - Wikipedia, the free encyclopedia might help. This same kind of controller is used in many applications, for example to control heating elements in a water heater or an oven. No computers required.

 

[jzzsxm]

I shot a quick little video to show it steering past the center of the lane when changing lanes.

I know that graph is self explanatory to a lot folks here, but a little explanation for the rest of us would help. Like perhaps, I am sure 'c' is the speed of light, right?

Imagine the Tesla driving from the left side of the graph to the right and following the path of the purple line. Pretend the horizontal dashed lines are the lane. That's what's going on. Not quite as extreme, but basically the Tesla overshoots the center of the lane and then comes back to it.

 

[Electroman]

Thank you @jzzsxm . Your video and the explanation now reminds of several times that this has happened. It never overshot the lane, but sure it went a little too much off-center and then slowly corrected itself to the center. And this behaviour was a little nerve wracking initially when there is a car or truck right on the lane adjacent to the target lane because you are always worried is it going to overshoot. But now that I have more confidence in the system, I am not worried but I am very watchful.

 

[DoubleE]

I shot a quick little video to show it steering past the center of the lane when changing lanes.

Imagine the Tesla driving from the left side of the graph to the right and following the path of the purple line. Pretend the horizontal dashed lines are the lane. That's what's going on. Not quite as extreme, but basically the Tesla overshoots the center of the lane and then comes back to it.

Exactly what mine does too. And also when there is another car in the neighboring lane.

 

[RDoc]

I'm going to do the obvious and go out and deliberately try it. The hypothesis is that when changing lanes, if there is another vehicle in the adjacent lane to the targeted lane, the car will overshoot. If there isn't one, it won't.

I'll get back with my results.

 

[3Victoria]

Just a suggestion. If the turn signal is held on too long then does the car think you want to change to the next lane (ie two lanes from the original)?

 

[CHGolferJim]

In casual observation on a few occasions, just holding the lever down but not far enough for it to catch, the lane change seems smoother. Sometimes if I just pushed the lever down to catch, the Lane change was a little abrupt. Assume this will be smoothed in later version.

 

[Larry Hutchinson]

I've now observed a slight overshoot of the center myself (traded in my old S40 for a 70 almost 2 weeks ago)

 

[RDoc]

Just a suggestion. If the turn signal is held on too long then does the car think you want to change to the next lane (ie two lanes from the original)?

no it only goes over 1 lane