TACC with 7.0 - unexpected slowdowns

[mwulff]

@LetsGoFast: I refer to built-in navigation (not the google maps view) as "navigon" because that is what Tesla is using for their navigation as far as I know.

Navigon is in no way a brand name in Europe. I bet most people have never heard of them.

The funny thing is that the google maps display is correct and the satnav in the car is wrong/outdated. Leads to some funny situations.

But yes, the problem occurs whenever the outdated navigon-maps thinks there are small roads when in reality there is a highspeed bypass.

This is quite disturbing because you could be driving along a new highway at 130 km/h and suddenly the car breaks because that particular highway used to be a small trail. In close traffic it could be really dangerous and it happens without any warnings what so ever.

- - - Updated - - -

I really wish Tesla would give us some kind of diagnostic screen to let us know what in the world it's doing sometimes.

There have been reports that the TACC slows down for corners so maybe it's expecting a corner.

It's expecting something that is not there that's for sure. Really scary to have the car apply the brakes like that.

 

[mesalum]

I too have the same problem here in No. California. After seeing this thread I realized the issue is probably the new roadway. The roadway had a 90 degree turn that was "smoothed" out. You can see the photo I attached. To the left of the words 'Grand Line Road' you see the 90 degree turn. The old Grand Line Road included that turn. When I enter the new portion the car slows down. I have not let it get to the slowest as I take over. Next time I will let it and see when it starts to speed back up. This happens both with TACC and autopilot. There are no stopped cars along the road side. The lanes are very nicely marked since it's new construction. There have been no cars around me. The road way is two lanes each direction with a raised mound as the divider. I am not using navigation as I know my way home. Also this happens both directions when I enter that new area. Next time I take the Tesla to work I will use navigation and see what it thinks about the road. The google navigation knows about the new road.

 

[WarrenInCA]

Some readers will know Highway 17 between Scotts Valley and Los Gatos. My car has been slowing for the curves since the firmware update. I though it had to do with the car "learning".

 

[CHG-ON]

I live in Ben Lomond. Hi Neighbor! Mine also slows when entering curves. I actually like it. It may be a bit slower than my style. But I like the safety aspect of it.

I think the issue here is definitely related to what the internal maps are thinking. But I am surprised that it is that intelligent. I have not seen any info. saying that the car will slow based on its perceived speed limits. Lots of rumours, but no facts that they have actually released it. My car has slowed for curves on country roads since about a month or so after TACC was released. I have not noticed it on a freeway, except for 17, which is quite curvy.

Some readers will know Highway 17 between Scotts Valley and Los Gatos. My car has been slowing for the curves since the firmware update. I though it had to do with the car "learning".

 

[evp]

I've had the same situation and can reproduce it pretty easily. It happens to me on overpasses. One is a new overpass which isn't even on the GPS map yet. My theory is that it speeds back up after it has crossed all the lanes of the road beneath.

I'm driving a P85 loaner today while my drive unit on my "classic" S85 gets replaced, so today is my first experience with TACC and Lane-following. I noticed exactly the same slow-down and speed-up. Couldn't figure out the cause, but looking back its was when I was on an overpass crossing the Interstate.

 

[mwulff]

I just had a response from Tesla that said: "We have forwarded this to your local service center". As if they can fix software bugs. Also tried driving the same stretch with the autopilot disabled (turned off in settings) and still the problem persists.

 

[stevezzzz]

Oh, good: I'm not alone. I'm seeing very similar behavior on a particular stretch of new roadway near my house. It's well marked, with only a gentle, sweeping curve and a slight uphill. Before 7.0, TACC had no trouble maintaining a constant speed through the stretch; with 7.0, the speed varies constantly, in herky-jerky fashion (it doesn't just slow down, it hesitates and speeds back up, several times, before slowing down to the point I have to take control). Watching the display, it's clear that Autopilot knows perfectly well where the lane is and steers quite smoothly; the erratic speed behavior is repeatable, and occurs whether or not auto-steer is engaged.

 

[brkaus]

Interesting. Looks like they are using navigon as the dataset to build their models for the auto driving. Makes sense as the navigon data is available offline and Google is not.

They might even be getting a super deal in navigon data in trade for providing updates back to them. Something Tesla may not be too excited to do with Google.

Does lead to strange behavior though when what you see us not what you get.

 

[steph280]

I experienced the slowdown this last Sunday on the 5 freeway south in Norwalk, southern california. It was in the middle of freeway with lots of cars, but all were moving at 65-70mph with safe distance. There were construction in the area and I didn't think much of it, just disengaged TACC and took over as the cars behind me started tailgating.

 

[WoodyOHill]

Exactly what version is your car on? v7.0 (2.7.x) what's the x?

 

[Coiled]

I've experienced this and I have a couple of different hypotheses to add to the mix:

1. My car also slows down at a specific spot on a nearby road. It's perfectly straight and there's nothing there that would cause a slowdown, but it speeds up immediately after I pass a large pole on my right. That made me suspect that it might have something to do with solid objects on the side of the road messing with the radar. In OP's video, at least for some slowdowns it looks like there are also objects on the roadside (signs at bumper level on both sides at the beginning of a bridge, for example) that could be causing a radar glitch.

2. Could this be the downside of fleet learning? Elon did say that the car would learn the safe speed for a particular curve or stretch of road. Maybe somebody (or multiple people) had to slow down in those spots due to traffic or whatever, and the car incorrectly learned that as the safe speed there. In that case, more statistics from more drivers who are not using TACC on that route should eventually cause the car to "unlearn" the wrong speed.

As for some of the other interesting hypotheses mentioned above: TACC may slow down for curves but I don't think it does for an expected turn onto a different street (due to nav routing) or a changing speed limit, so those should not be issues. In my case, I don't think the road is unknown to nav so that shouldn't be the cause, but I will have to check it out. I guess I will know if nav tries to route me around that section.

Of course there could be different causes in different situations, which makes it much more difficult to analyze.

 

[Coiled]

1. My car also slows down at a specific spot on a nearby road. It's perfectly straight and there's nothing there that would cause a slowdown, but it speeds up immediately after I pass a large pole on my right. That made me suspect that it might have something to do with solid objects on the side of the road messing with the radar. In OP's video, at least for some slowdowns it looks like there are also objects on the roadside (signs at bumper level on both sides at the beginning of a bridge, for example) that could be causing a radar glitch.

To clarify: The radar glitch could be due to TACC's increased sensitivity to stopped cars. It could be that some arrangement of roadside objects returns a radar and/or visual signature that looks like a stopped car. This could be the kind of problem they were originally trying to avoid when they made TACC less sensitive to stationary objects.

 

[ElectricTundra]

I've had this as well. The most consistent place is a new bit of motorway (35E just north of St Paul) crossing over some train tracks and a street. The old bit was straight and this new bit is as well so I don't think it's a curve issue unless it's thinking it's on a road that had been below here (not sure if such a road existed though).

 

[!S4]

To clarify: The radar glitch could be due to TACC's increased sensitivity to stopped cars. It could be that some arrangement of roadside objects returns a radar and/or visual signature that looks like a stopped car. This could be the kind of problem they were originally trying to avoid when they made TACC less sensitive to stationary objects.

In my experience it has always happened on highways with no cars stopped on the shoulders. No difference if there are any cars around and I've watched the screen and it does not indicate detecting any cars. Given the correlation with overpasses, it seems more likely to be GPS data related.

 

[Max*]

Happened again. TACC set to 30mph in a 25mph zone (city, blah blah, I know), sharp turn ahead, single lane, no cars, no parked cars, no oncoming traffic.

Car started to slow down before the turn to 28mph, then probably about 25mph in the turn, and then back up to 30mph after the turn.

 

[mesalum]

Happened again for me too. Today had TACC set (no autopilot) at 60mph on a 55mph roadway. Car slowed to 34mph for some time and I let it go this time. Finally went back to the 60mph. I was not following any route set through navigation. No other vehicles around me, I was alone on the road.

I have V7.0 - 2.7.56.

 

[green1]

I had this happen once, middle of nowhere on a 4 lane divided highway with no traffic around, car suddenly dropped to 50km/hr from 110km/hr then resumed. My initial suspicion was that the car had mistaken some shadow or other item for a car and slowed appropriately, howver what I found really interesting was that at no time did the display indicate any vehicle in front, or any sensor arcs on the sides, it also showed good clear lane markings left and right.

The most likely situation seems to be as mentioned up thread that the car is doing it's "slowing for a curve" thing, even though no curve is present (thinks there's a curve based on... map data maybe?) The section I had it happen on was, as several others have mentioned, a fairly new piece of road, where the old road may have had a sharp curve.

 

[R²B]

This definitely has nothing to do with a curve. Happens to me everyday in the exact same place on a brand new freeway that the navigon maps don't know about yet. It's a GPS maps thing.

 

[green1]

This definitely has nothing to do with a curve. Happens to me everyday in the exact same place on a brand new freeway that the navigon maps don't know about yet. It's a GPS maps thing.

It could be the curve algorithm messing it up though based on lack of information. The reports mostly seem to be from places where the navigation maps don't know the road is straight.

 

[mwulff]

I just got a confirmation from Tesla that it is indeed that gps/outdated navigon maps that cause these slowdowns. That explains why it happens in the same place everywhere. My video and report has been forwarded to the bug-system/dev team.