Autopilot appears to turn car against incoming car

[gsxdsm]

Except auto steer DOES veer into the oncoming lane given that terrain. The car applies significant force turning the wheel to the left. I have experienced it, and can replicate it on a similar road.

In the video, AutoPilot is off when the wheel turns.

In your case, is autopilot still on?

 

[mmccord]

In the video, AutoPilot is off when the wheel turns.

In your case, is autopilot still on?

Yes, although once it crosses over the double yellow far enough, it usually shuts off with the red alert. Sometimes it recovers on its own, but not before going halfway into the other lane.

 

[JonathanD]

I'm sort of surprised Tesla didn't have autopilot use restricted by GPS to larger multi-lane roads.

 

[mobe]

Autosteer applies or the terrain applies and you don't have hands on the wheel? Seems like terrain to me. The autosteer light is off.

Check out the slo mo video that was posted. The terrain pulled the car to the left, the autopilot tried to apply torque to the right but was overcome and disconnected. The instant that it STOPPED applying torque to the right, the car veered to the left.

 

[mmccord]

I don't see anything in the terrain that would push the car to the left. Cars tend to drive (relatively) straight. They don't veer off like that on their own.

 

[gregincal]

I don't see anything in the terrain that would push the car to the left. Cars tend to drive (relatively) straight. They don't veer off like that on their own.

While I agree that I don't see why the car would veer like that, it is definitely true that the wheel starts to turn the moment after auto steer disengages. Therefore it doesn't seem reasonable to me that auto steer could be making the turn when it has clearly disengaged.

 

[mmccord]

While I agree that I don't see why the car would veer like that, it is definitely true that the wheel starts to turn the moment after auto steer disengages. Therefore it doesn't seem reasonable to me that auto steer could be making the turn when it has clearly disengaged.

Maybe autosteer doesn't really shut off until it detects you have actually grabbed the wheel, so it's doing it's best guess even with low certainty? Hard to say for sure. I haven't had the cojones to look at my dash while AP is steering me into the other lane, so I don't know the status of the indicator lights at the time.
I guess I could try to mount my dashcam in the back of the car to find out, or find a volunteer to sit back there and hold it.

 

[eloder]

I don't see anything in the terrain that would push the car to the left. Cars tend to drive (relatively) straight. They don't veer off like that on their own.

Nothing caused it to veer like that except for... oh, physics and intertia from coming out of a turn.

It's really really easy to recreate what happened in the video. Take a turn, and at the end of the turn let go of the wheel. Watch what happens to your car.

 

[mmccord]

I bet i can do it on a straight road

 

[gsxdsm]

I bet i can do it on a straight road

I bet you can't.

 

[schonelucht]

Drive normally on a straight road that has a sharp turn to the right. Enter the turn at a normal speed, then turn the steering wheel to follow the road. At the apex of the turn, remove your hands from the wheel.

You will see that the forward inertia of the car will cause the wheel to turn back to the left and the car will veer out of the turn. When the force of your hands on the wheel turning it to the right is removed, an equal but OPPOSITE force is applied to the wheel to the left.

Except that in the video the car is not in a turn and that the turn before and after are all to the left with the car veering to the inside of the turn (ie, left also). Centrifugal forces are most certainly not at play here.

 

[AWDtsla]

Except that in the video the car is not in a turn and that the turn before and after are all to the left with the car veering to the inside of the turn (ie, left also). Centrifugal forces are most certainly not at play here.

Simple left bank is all it takes. The road had lots of elevation changes and bank angle changes. The only indication of bank angle in the video is to use the direction of the trees, you can see the angle between the road the and trees on the left hand side is much smaller than 90 degrees.

This is what happens when _no one_ is steering the car.

edit - and after watching that over and over several times I have to say that guy is a moron. The car keeps losing track of the lines, ping pongs between them, then you can even hear the lane departure warning go off as the car crosses the white line on the right side (This must be a separate system). He still keeps his hands off the wheel until the car just gives up. I'm surprised he didn't try dangling weights from the wheel, it would have done a better job steering that he did.

- - - Updated - - -

Until they "build a better idiot" autopilot should do steering angle hold when disengaging. It can use input torque to disable the hold. However last steering angle is likely to be the best guess.

 

[Matias]

Until they "build a better idiot" autopilot should do steering angle hold when disengaging. It can use input torque to disable the hold. However last steering angle is likely to be the best guess.

Or maybe change the steering angle slowly to neutral.

 

[ev-enthusiast]

A Model S is currently driving coast to coast on autopilot: twitter

 

[supratachophobia]

Currently experienced this twice today. The 1st was actually a similar situation with following a car, he drifts 2ft to the right while staying in the same lane, car locks onto oncoming semi-truck instead and attempts to go left of center. This was a two lane road.

Now the highway was uneventful in a good way. I think we went a total of 75 miles @ 75mph without a single steering intervention whatsoever, only lane changes. That was pretty impressive. I eventually started just holding my hands up in a pretend mid-air steering wheel and making eye-contact with people on the left and right when I wasn't resting my palm on the lower part.