Autopilot Following Distance Changes near Exits

[jsmay311]

I noticed an interesting Autopilot behavior recently and I couldn't find it mentioned in any other threads, so I figured I'd post it here and share my observations.

While I was...

1. driving on the highway/expressway
2. in Autopilot (not NOA, although idk if that matters)
3. in the right lane
4. following a vehicle in front of me that was limiting my speed
5. with a AP following distance of "1"

...any time there was an upcoming exit ramp on the right side of the expressway, starting roughly a third of a mile before the exit the car would start gradually slowing down and increasing the following distance with the vehicle ahead of me until the following distance was 50-100% longer than normal. Then as soon as I had passed the exit, the car would speed back up until it had narrowed the following distance back to the "normal" distance.

While this behavior was not super obvious or obtrusive, I did find the repetitive and unnecessary decelerations/accelerations quite annoying after a while.

The only likely explanation I can think of for why Tesla would have programmed this behavior into EAP is that it is intended to reduce the frequency and severity of another annoying EAP behavior caused when the car in front of you takes an exit and starts slowing down a bit before it is 100% clear of the right lane. When this happens, EAP often overreacts and slows way down (when no human driver would since the human would recognize that the car is leaving the lane), and then as soon as the lead vehicle is perceived by EAP as being 100% out of the lane, EAP overreacts again and and abruptly zooms back up to the target speed in a jerky/unpleasant manner.

Assuming this is in fact the reason for this particular behavior, it sure would've been nice if Tesla could have addressed this original issue *intelligently* by teaching EAP to recognize when a lead car is taking an exit so it can maintain its speed in those situations, instead of slapping this "band-aid" fix on it. They're really just replacing one annoying, dumb deficiency with another also-annoying and also-dumb "fix".

 

[ewoodrick]

Doesn't happen to me. What mode do you have the autopilot in? Maybe make it a little less aggressive.

 

[JeffK]

I can confirm this happens to me when on NoA.

It goes from freeway speeds, slows down for no reason as the car in front of me gains distance, as it gets closer to the turn it speeds up then switches to normal autopilot to handle the rest of the exit.
(fast... slow.... fast, then slower for the turn)

I should note the section that it's going super slow on is straight and not a curve, but in the exit lane.

 

[jsmay311]

Doesn't happen to me. What mode do you have the autopilot in? Maybe make it a little less aggressive.

I don’t know what you mean by “mode” or “less aggressive”.

I already stated in the OP that I was not using Navigate on Autopilot (so the “Mad Max” setting is irrelevant) and the following dinstance was set to 1.

 

[jsmay311]

I can confirm this happens to me when on NoA.

It goes from freeway speeds, slows down for no reason as the car in front of me gains distance, as it gets closer to the turn it speeds up then switches to normal autopilot to handle the rest of the exit.
(fast... slow.... fast, then slower for the turn)

I should note the section that it's going super slow on is straight and not a curve, but in the exit lane.

To be clear, I was talking about situations where I’m driving past an exit in the right lane, not taking an exit. And I wasn’t using NoA.

I would think that the programmed behavior for adjusting speeds and following distances when taking an exit with NoA enabled is likely to be a distinctn set of algorithm than when staying in the right lane with NoA off.

 

[ecarfan]

any time there was an upcoming exit ramp on the right side of the expressway, starting roughly a third of a mile before the exit the car would start gradually slowing down and increasing the following distance with the vehicle ahead of me until the following distance was 50-100% longer than normal. Then as soon as I had passed the exit, the car would speed back up until it had narrowed the following distance back to the "normal" distance.

A possible explanation for that behavior is that it is planning in advance for the possibility of cars merging into your lane from the on ramp that is coming up just after the off ramp (exit) you are driving by.

In the past few months, EAP and NAP have improved significantly in regards to anticipating and accommodating merging cars when I am in the far right lane on a freeway/limited access highway. It’s really been quite amazing to see how the car detects a merging vehicle in the on ramp lane and then either slows slightly if it predicts that vebhicle will merge ahead of me, doesn’t change speed if it predicts it won’t merge close to me, or even accelerates slightly to get ahead of the merging vehicle.

Of course it’s not perfect and in crowded traffic conditions it doesn’t always make the correct decision. But earlier this year the car had zero capability for handling merging cars. Now it often handles them correctly.

 

[jsmay311]

A possible explanation for that behavior is that it is planning in advance for the possibility of cars merging into your lane from the on ramp that is coming up just after the off ramp (exit) you are driving by.

In the past few months, EAP and NAP have improved significantly in regards to anticipating and accommodating merging cars when I am in the far right lane on a freeway/limited access highway. It’s really been quite amazing to see how the car detects a merging vehicle in the on ramp lane and then either slows slightly if it predicts that vebhicle will merge ahead of me, doesn’t change speed if it predicts it won’t merge close to me, or even accelerates slightly to get ahead of the merging vehicle.

Nope. It very clearly only happens when going past an exit / off-ramp.

Right after passing the exit, the car speeds back up and starts closing the following distance and is back to the normal following distance well before approaching any subsequent on-ramp (assuming both ramps coming from/to the same street).

I also haven’t noticed any advanced behavior like you’re describing in anticipating merging vehicles. Maybe it’s because I usually don’t trust it to handle those situations and disengage AP when cars are about to merge into my lane. Maybe I’ll have to do some tests — but it’s hard to do so without risking driving like an A-hole.

 

[JeffK]

Nope. It very clearly only happens when going past an exit / off-ramp.

This is likely due to GPS inaccuracies. Does it still do it if you get in the lane(s) to the left?

 

[David L]

I can confirm that this happens to me on AP without NoA enabled. It's most noticeable in the leftmost lane when there's an entrance to a HOV lane. I've even disabled HOV lanes as an option from the navigation and it still slows down. I find this behavior to be extremely dangerous, especially in high traffic situations in LA, when the drivers behind me get mad that I'm slowing down for no reason and follow even more closely. I usually have to push on the accelerator until I pass the HOV entrance.

 

[mtndrew1]

This happens on my car and drives me nuts. It should be reserved for NoA drives so as to allow for easier merging by the computer. I hate it when just using regular autopilot.

 

[ConcordeSST]

I wonder if this explains why my follow distance setting doesn't seem to work right when I'm on autopilot. I typically drive in the right lane and find that a one car spacing setting usually results in two or more car spacing. I never paid attention if I was driving past an exit.

 

[ecarfan]

I also haven’t noticed any advanced behavior like you’re describing in anticipating merging vehicles. Maybe it’s because I usually don’t trust it to handle those situations and disengage AP when cars are about to merge into my lane. Maybe I’ll have to do some tests — but it’s hard to do so without risking driving like an A-hole.

I haven’t had that problem.

 

[mswlogo]

Sounds like it's being cautious about the car ahead possibly taking the exit especially when following at the closest range.

I'd be curious what it does when set to various distances.

 

[jsmay311]

I wonder if this explains why my follow distance setting doesn't seem to work right when I'm on autopilot. I typically drive in the right lane and find that a one car spacing setting usually results in two or more car spacing.

The following distance number does not equate directly to the number of car lengths.

The actual following distance is variable based on speed. In my experience, when going ~60mph a “1” following distance results in being maybe 4 car lengths back (just a rough estimate from my recollection).

 

[jsmay311]

Nope. It very clearly only happens when going past an exit / off-ramp.

This is likely due to GPS inaccuracies. Does it still do it if you get in the lane(s) to the left?

Of course it isn’t due to GPS inaccuracies. That makes no sense. The on-ramps and exit ramps for a given interchange are separated by hundreds of yards -- far too great a distance for GPS inaccuracies to play any role.

And, no, I haven’t observed it happening anywhere other than the right lane -- although I suspect it would happen in the left lane if there was an exit on the left. And, granted, I just recently noticed it, so I can’t say with 100% confidence that it wouldn’t happen in another lane. I’ll keep an eye out for it in the future and update if it happens in any lane NOT adjacent to an exit ramp.

 

[JeffK]

Of course it isn’t due to GPS inaccuracies. That makes no sense.

Makes perfect sense, especially if you've ever used consumer grade GPS for navigation. Sometimes these devices temporarily think you're in the next lane over, especially near exit ramps.

If the car thinks you're near the exit ramp it'll slow itself down. Get another lane over and it won't do anything like that because you've given it a larger margin.

I mean they even patented a method of improving the car's ability to figure out it's own position.
https://electrek.co/2018/12/09/tesl...te-gps-positioning-for-self-driving-vehicles/

Here's a quote from the patent:

For example, a smartphone with a positioning receiver may be able to determine its position to within five meters of the smartphone. The accuracy of the position determination may worsen when the receiver is in proximity of buildings, bridges, trees, or other structures.

but guess what, five meters is enough to put you in the next lane over as if you'd taken the exit. It will/does eventually correct itself, but by then it's already slowed down quite a bit as it does when you normally take the exit.

The on-ramps and exit ramps for a given interchange are separated by hundreds of yards -- far too great a distance for GPS inaccuracies to play any role.

This is irrelevant, if you're in the right lane and the exit is on the right then it's only one lane over not "hundreds of yards away"

 

[jsmay311]

Makes perfect sense, especially if you've ever used consumer grade GPS for navigation. Sometimes these devices temporarily think you're in the next lane over, especially near exit ramps.

If the car thinks you're near the exit ramp it'll slow itself down. Get another lane over and it won't do anything like that because you've given it a larger margin.

I mean they even patented a method of improving the car's ability to figure out it's own position.
Tesla patents technology for more accurate GPS positioning

Here's a quote from the patent:



but guess what, five meters is enough to put you in the next lane over as if you'd taken the exit. It will/does eventually correct itself, but by then it's already slowed down quite a bit as it does when you normally take the exit.



This is irrelevant, if you're in the right lane and the exit is on the right then it's only one lane over not "hundreds of yards away"

Sorry, I misunderstood. The comment of mine that you originally quoted was in response to speculation that the following distance was increasing to allow space for possible merging vehicles at upcoming on-ramps, so that's what I thought you were talking about.

Regardless, the GPS inaccuracies theory still doesn’t fit here because (as I mentioned in the OP) the following distance starts increasing many hundreds of yards before the exit, way before the fork for the exit ramp begins to take shape. I.E., it can't possibly think I'm in an exit lane that doesn't yet exist.
And it only happens when closely following a lead vehicle. If it was slowing down because it thought I was on an off-ramp, I would think it would slow down regardless of whether or not there was a vehicle in front of me.


And, a bit off-topic, but I always figured the car uses its camera(s) to determine which lane it's in, not GPS -- for the exact accuracy limitation reasons that you mentioned. Idk if that's true or not though. Or maybe it's a combination of GPS and camera.(?)

 

[nvx1977]

@jsmay311 I have noticed this same behavior. It was introduced with Navigate on Autopilot but it happens regardless if you have the NoA active or not. It's really annoying; I've been stepping on the accelerator to keep distance when I'm near an exit. Doesn't matter if that's my exit or not. On my commute stretch of highway, there are a lot of exits, so this often exhibits as the car not respecting my follow distance (1). And I've had EAP since March, so I'm very familiar with what the 1 setting does in a variety of speeds.

It happens from all lanes.

It feels like this feature was intended to be active for NoA drives, but they decided to incorporate it into all of EAP, maybe for safety considerations, with the assumption that there's lots of merging activity happening around exits (before and after) so why not leave more space to accommodate for this traffic. Kinda wish it was silo'd to just NoA though. I might leave that feedback for Tesla.

 

[camrunr]

Chiming in with "me too". Not sure when it started, but it was one of the recent updates. I'm positive this wasn't a thing for most of the 17k miles i have on my 3.

 

[jsmay311]

It happens from all lanes.

Not in my experience.

I just did some testing while driving on the expressway and it did not happen while going past multiple exits in the 2nd-from-the-right lane.