Cruise control inappropriately slamming on brakes

[sleepydoc]

So a big part of the problem is that what you want TACC to do and what Tesla wants TACC to do are quite different.
Tesla seems to want TACC to also:
adjust speed for speed limits
adjust speed for curves (and maybe sometimes hills?)
stop for red lights/stop signs
stop for pedestrians
stop for UFOs
slow for construction? (maybe just a side effect of barrels in the lane or no lane lines?)
[Probably many, many more things.]

I think 2019 was probably the peak for [dumb] TACC.

Yeah, I guess you’re right - I wan’t TACC to work like it’s designed to and as good as every other adaptive cruise control. Tesla’s ok with ‘fancy’ adaptive cruise that doesn’t work. Seriously, all that stuff is great, but it’s pointless if the system fails at its primary mission.

I’ve said this a thousand times, but I have routine slowdowns on straight roads in clear weather. Maybe TACC is imagining a curve or a stoplight-shaped UFO, but it doesn’t matter - if it is, the system is failing.

 

[JohnB007]

I’ve said this a thousand times, but I have routine slowdowns on straight roads in clear weather. Maybe TACC is imagining a curve or a stoplight-shaped UFO, but it doesn’t matter - if it is, the system is failing.

Are you able to see what could potentially be fooling TACC?

For me, TACC doesn't slow down at all now (I'm assuming this happens on highways), and when FSD slows down, I can usually (not 100%) see reasons why, which of course needs further tuning by Tesla.

 

[Knightshade]

Are you able to see what could potentially be fooling TACC?

For me, TACC doesn't slow down at all now (I'm assuming this happens on highways), and when FSD slows down, I can usually (not 100%) see reasons why, which of course needs further tuning by Tesla.

Yup... in tens of thousands of miles, and nearly 4 years- and MOST of that with some form of TACC, AP, or now FSDbeta applied, I've virtually never had actual "phantom" braking where no obvious source of it existed... there was always a visible cause, even if I disagreed with the choice to slow down for said cause.

The nearest ever was earlier Beta FSD versions where on one road when it was cresting a hill (and thus had no real visibility to what was beyond) it'd slow down maybe 5-6 mph which 'felt' harsh but wasn't nearly the accident-causing kinda thing a few here describe--- and multiple updates ago even that was smoothed out- it'll still slow 1-2 mph there now, but that's not really an issue at all.

 

[sleepydoc]

Are you able to see what could potentially be fooling TACC?

For me, TACC doesn't slow down at all now (I'm assuming this happens on highways), and when FSD slows down, I can usually (not 100%) see reasons why, which of course needs further tuning by Tesla.

Sometimes yes, often times no. In the end, does it matter? People seem to get caught up in overanalyzing the system and finding explanations for why it fails. Say there's a sign that's confusing it. Great - I have a reason but it still didn't function as intended.

 

[JohnB007]

Sometimes yes, often times no. In the end, does it matter? People seem to get caught up in overanalyzing the system and finding explanations for why it fails. Say there's a sign that's confusing it. Great - I have a reason but it still didn't function as intended.

I only asked because if there is a valid reason, I'm hoping Tesla will eventually figure it out given enough time.

 

[JohnB007]

Yup... in tens of thousands of miles, and nearly 4 years- and MOST of that with some form of TACC, AP, or now FSDbeta applied, I've virtually never had actual "phantom" braking where no obvious source of it existed... there was always a visible cause, even if I disagreed with the choice to slow down for said cause.

The nearest ever was earlier Beta FSD versions where on one road when it was cresting a hill (and thus had no real visibility to what was beyond) it'd slow down maybe 5-6 mph which 'felt' harsh but wasn't nearly the accident-causing kinda thing a few here describe--- and multiple updates ago even that was smoothed out- it'll still slow 1-2 mph there now, but that's not really an issue at all.

My personal experience is actually closer to yours than speedydoc's (but far less driven miles to compare against).

Don't know the reason, but perhaps I'm lucky that I live in a area that is more Tesla-friendly.

 

[sleepydoc]

Here's a good example: I was driving on the road pictured below using TACC only. Full lane, concrete median, 6-8 foot shoulder and a bike path on the side separated from the road by a strip of grass. There was a man riding his bike about where the arrow is in the picture. No traffic coming the opposite direction and a car traveling at the same speed about 2-300 feet ahead of me. The car randomly slowed down 5-8 MPH, then sped back up again.

The only possible reason I could see was the cyclist but he was about as non-threatening as you can get short of being behind a wall. Was it dangerous? absolutely not. Necessary? Not in the least. Annoying? Yes, although I've gotten pretty used to it by now.

another facet is Tesla's focus on self driving. If you're making a level-2, driver assist feature you expect the driver to be paying attention and in control. You don't want adaptive cruise to slam into a car in front of you, but if it misses a cyclist on the shoulder it's not a failure, per se. With Tesla's (supposed) push towards level 5 autonomy where the car is in charge of everything, the tolerance for mistakes is far lower. In the former, the system is helping the driver and catching things s/he misses. In the latter it's responsible for everything and a miss is a much bigger deal. Still not really an excuse, but another possible explanation.

I'm hoping Tesla will eventually figure it out given enough time.

Ditto!

 

[DrChaos]

Here's a good example: I was driving on the road pictured below using TACC only. Full lane, concrete median, 6-8 foot shoulder and a bike path on the side separated from the road by a strip of grass. There was a man riding his bike about where the arrow is in the picture. No traffic coming the opposite direction and a car traveling at the same speed about 2-300 feet ahead of me. The car randomly slowed down 5-8 MPH, then sped back up again.

The only possible reason I could see was the cyclist but he was about as non-threatening as you can get short of being behind a wall. Was it dangerous? absolutely not. Necessary? Not in the least. Annoying? Yes, although I've gotten pretty used to it by now.

another facet is Tesla's focus on self driving. If you're making a level-2, driver assist feature you expect the driver to be paying attention and in control. You don't want adaptive cruise to slam into a car in front of you, but if it misses a cyclist on the shoulder it's not a failure, per se. With Tesla's (supposed) push towards level 5 autonomy where the car is in charge of everything, the tolerance for mistakes is far lower. In the former, the system is helping the driver and catching things s/he misses. In the latter it's responsible for everything and a miss is a much bigger deal. Still not really an excuse, but another possible explanation.


Ditto!

View attachment 850301

There is a chance it might have nothing to do with the cyclist, but the upcoming dip/decline in the road.

I think there is some 'phantom braking' frequently when there are elevation undulations and it's harder for the camera to see the path ahead. Most likely it has a calculation of "how many seconds ahead can I see at my current speed" and if that is too small, it slows down.

Or perhaps the cyclist made some move which was interpreted as possibly turning towards the drive lane, possibly erroneously.

It's also possible the combination of the various observations in a neural network puts it over the threshold of 'slow down' where one wouldn't alone. In a machine learning system it's usually not feasible to think of the output behavior as a fixed combination of cognitively understandable logical rules. It's a quantitative weighting and it can have unobvious or inexplicable behavior some of the time. Humans are like that too, we confabulate linguistic explanations post hoc, but that isn't what really drove the instant decision.

 

[Kevy Baby]

Yeah, I guess you’re right - I wan’t TACC to work like it’s designed to and as good as every other adaptive cruise control. Tesla’s ok with ‘fancy’ adaptive cruise that doesn’t work. Seriously, all that stuff is great, but it’s pointless if the system fails at its primary mission.

I’ve said this a thousand times, but I have routine slowdowns on straight roads in clear weather. Maybe TACC is imagining a curve or a stoplight-shaped UFO, but it doesn’t matter - if it is, the system is failing.

You and I must be using different versions of TACC because it has worked pretty well since I got my car three years ago and it continues to get better. It's not perfect, with occasional slow-downs (but always in the same place, but I cannot figure out what is triggering it). The annoying bug of adjusting speed (down) after some freeway interchanges seems to have been corrected. Phantom braking is a thing of the past.

I wouldn't drive without it.

 

[DrChaos]

You and I must be using different versions of TACC because it has worked pretty well since I got my car three years ago and it continues to get better. It's not perfect, with occasional slow-downs (but always in the same place, but I cannot figure out what is triggering it). The annoying bug of adjusting speed (down) after some freeway interchanges seems to have been corrected. Phantom braking is a thing of the past.

I wouldn't drive without it.

I think it works significantly better in California than other places.

 

[Knightshade]

I think it works significantly better in California than other places.

Has been working fine here on the east coast for years too.

 

[JohnB007]

Works well in Ontario as well.

 

[Kevy Baby]

Maybe it just doesn't work in Minneapolis?

 

[sleepydoc]

Maybe it just doesn't work in Minneapolis?

I've had it do it in Wisconsin as well...

 

[alexgr]

Has been working fine here on the east coast for years too.

It looks as the AP is trained better for certain roads. I had multiple phantom breaking events on I-80 in Wyoming even though I could drive 300+ miles in some other states without a single issue.

 

[sleepydoc]

It looks as the AP is trained better for certain roads. I had multiple phantom breaking events on I-80 in Wyoming even though I could drive 300+ miles in some other states without a single issue.

The inconsistency is frustrating. I’ve driven long periods on occasion without issue, then other times I can barely go more than a few miles. Some spots are repeatable, some spots are random. Interstates tend to be better, but I’ve had issues on them, too.

For all it’s inconsistencies, I‘ve never had it not slow down for a car, and when I drive our Odyssey with Dumb cruise I realize I’d still rather have a flawed ‘smart’ cruise than dumb cruise!

In general it has been slowly improving. I’m hoping 2022.20 will give further improvements - if I ever get it!

 

[terranx]

you may not appreciate this, but TACC is paying attention to every 360 threat potentially around your car. Your “better” old cars only concerned themselves with following distance to whatever was directly in front of them.

I can pay attention to my surroundings just fine. If the car is worse than me, I don’t want it to attempt it. Just let me set cruise and let me worry about the surroundings.

 

[Kevy Baby]

If the car is worse than me...

It isn't. No matter how good you think you are, the car driving is infinitely safer than any driver out there.

 

[terranx]

It isn't. No matter how good you think you are, the car driving is infinitely safer than any driver out there.

If that’s truly how you feel, I fear for the people you share the road with.

 

[Kevy Baby]

I figured it would stir some people up. The truth can hurt.