Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

8.0 (2.50.185) caution using TACC/Autosteer features

This site may earn commission on affiliate links.
I sure hope that Tesla has a validation engineer assigned to follow these threads like a hawk!

Even if they are monitoring the cars' telemetry directly, they need to know the human perceptions. AP is semi-autonomy; the human driver is in the loop. Tesla needs to be aware of perception as well as physics.
 
I believe that with AP 1.0 the car doesn't brake when someone cuts in front of me that is traveling faster than I am. If so, the description of 2.0 works like 1.0.

As for the phantom red car collision avoidance, I'd take the car in for service and make sure the RADAR is properly aligned.
 
  • Helpful
Reactions: 3Victoria
One frustrating thing about Autopilot is that when it does something you think it shouldn't, it could be a firmware bug, an anomalous road situation, a hardware fault or a wrong expectation. On a forum it's hard to know which it is when someone reports a problem. Having been in the software business, I know that some people are better at reporting problems than others. Also, you can't just generalize any problem to the entire fleet in all situations.

Reviewing my own dashcam footage, I've found that things really didn't happen the way I remembered them. I encourage folks to record anomalies and post footage, so that more experienced eyes (and Tesla) can look at them. I think the "Bug Report" voice command is very appropriate if the car brakes for a phantom "red car" ahead; the report commands take a snapshot of car systems and this can help Tesla diagnose the problem, beit hardware or software.
 
This is a great discussion. It reminds of earlier threads when AP1 was first released but this is now a much more complicated discussion. We have a new version of AP1 (8.0) and also AP2 being introduced for the first time. We have experienced AP1 drivers, new AP2 drivers and some who are perhaps have experienced both commenting on experience. I personally have been using AP1 since May and have about 13k miles of overall experience. I think it will be helpful for people weighing in/commenting to be more precise about their own AP experience level.

@ZeroDarkSilver described exactly how I think about and use AP1 (7 and 8) and I am extremely satisfied. I have tried it on un-divided highways and found it way too stressful to even consider - unless I already know the road really well.

As for 7.1 vs 8.0 AP1, I was initially quite disappointed with AP1 8.0 but over time I have come to love it. It does a much better job than AP1 7.0 at pacing, lane changes and overall driving experience. I am comfortable using it in heavy traffic - but increase my vigilance a few notches in those scenarios. I still don't consider using it on undivided highways. My impression is that AP 1 8.0 has improved significantly from its first release.

I personally am bummed that I won't have AP2 - but I have had a fantastic time with my X so I wouldn't trade that experience for AP2.
 
  • Like
Reactions: 3Victoria and BerTX
Serious question:
Does Tesla have employees or volunteers to drive these vehicles, in real time parallel to customer/owners, strictly to get the software up to speed and safe?
And also:
It seems to me that post #1 information would do greater good to all concerned if it were placed in the hands of the people who are actually developing these features. Public displays like this thread seem counter intuitive/productive to me....

The first priority is to keep drivers safe. I have personally learned a lot from others feedback and those who follow the thread will have better expectations on what AP does and does not do. I have reported to Tesla but the more eyes on the issues the better.

MXWing, I'm curious is this your first Tesla with Autopilot, or did you have a Tesla with AP1 before this? Just wondering if your commentary is comparing AP1 to AP2 or whether AP2 is a completely new experience for you.

I did not own a Tesla prior. Just saw AP1 in a test drive environment. I think coming from the perspective of what I would expect how things should work would be productive feedback to Tesla and fellow new owners. For example, no one would expect that TACC is beautiful when dealing with cars in motion but you would slam into an immobile object at 70mph is a bad but an assumption I could easily see other drivers making.

Posted this in another thread but applicable here too. I have model s though.

So something interesting: I had the update installed New Year's Eve. Took it out for a test drive New Year's Day, couldn't use anything, "camera calibrating" message. I drove max 20 miles New Year's Day. I drove it this morning, and was able to use TACC right away. So it does seem it's not necessarily a time driven in your car or miles driven in your car as far as I can tell.

I tested TACC and autosteer. I had to go out and find traffic on the freeway to test it, it won't work unless you are on the freeway and going under 35 mph. The one day in Southern California traffic is light is today lol, so I was only able to find about a 5 minute backup. But Autosteer worked great, but it wasn't a long test of a difficult freeway (5 freeway).

TACC worked everywhere I tried it. I drove about an hour on and off freeways with it. Over all on the freeway it works great, though it struggles when the radar hits a freeway overpass. It will slow down pretty dramatically for about 2 seconds then it will work itself out and continue on. I let it do this many times, it always recovered I would say It went from 65 down to about 45-50 mph. Not optimal, but I know tesla has talked about this issue and the past and specifically said fleet learning will address that issue. It didn't have that issue over most overpasses I would say, but it did happen 3-4 times.

One thing that worked great was merging onto a different freeway on ramp with a curve. TACC was set to 75, there was a broad curve to this onramp. TACC slowed down to 55-60 to handle the curve then speed back up just like I would have done. Smooth overall.

On the roads, I used it in heavy stop and go traffic, it worked decent most of the time. One time as i was driving next to a bicyclist in the bicycle lane it freaked out and slowed way down then recovered about a second later. The acceleration and breaking can be smoother for sure, but it's not horrible. I got used to it in about 10 minutes.

I think we have to remember Tesla is essentially starting over developing their own image recognition, sensor network etc. I like the idea that it's in house now, not having to depend on mobileeye. But with good comes bad, this means we have to go through all of this "beta" period crap where the software is not 100%. I have had the car about a month and about every two weeks I have recived new updates giving me new features. That's pretty awesome when you think about it. So, things are moving, just be patient.

Thank you so much for your feedback. Please keep sharing them. We need more data points as this is a critical juncture with first time Software in AP2. Regarding calibration, it seems it can be as low as 20 miles. But the kicker is the car has to be set on Park first after the calibration to actually turn in. It would be nice to be told this.

Also,TACC took a curve rated for 50 at 65. No slowdown from the car. I wonder if slowdown happened due to a car in front of you?

TACC slows down only if there is a vehicle in front, did it used to slow down on curves on AP1 as well? on its own? while you navigate the steering?

I am skeptical about this portion of the report. Right now, I am not believing in AP2 being able to read road signs well enough (if it actually does it at all?????) to slow down around curves. I'm not going to take a curve rated for 35 at 70 with TACC on to find out. :) If I see an opportunity like a speed limit of 40 and a curve at 25, I'll see it deaccelerates.

From the little I saw, I cannot confirm those findings that TACC drops speed from factors other than a moving object in front of it.

I believe that with AP 1.0 the car doesn't brake when someone cuts in front of me that is traveling faster than I am. If so, the description of 2.0 works like 1.0.

As for the phantom red car collision avoidance, I'd take the car in for service and make sure the RADAR is properly aligned.

Does the vehicle have its own tools to determine that Radar is not working correctly? I can consider service but would rather avoid that since I don't want to unnecessarily take away 'shop cycles' from other drivers who have more pressing issues.

One frustrating thing about Autopilot is that when it does something you think it shouldn't, it could be a firmware bug, an anomalous road situation, a hardware fault or a wrong expectation. On a forum it's hard to know which it is when someone reports a problem. Having been in the software business, I know that some people are better at reporting problems than others. Also, you can't just generalize any problem to the entire fleet in all situations.

Reviewing my own dashcam footage, I've found that things really didn't happen the way I remembered them. I encourage folks to record anomalies and post footage, so that more experienced eyes (and Tesla) can look at them. I think the "Bug Report" voice command is very appropriate if the car brakes for a phantom "red car" ahead; the report commands take a snapshot of car systems and this can help Tesla diagnose the problem, beit hardware or software.

I agree 100%. I'm in systems engineering so I get all sorts of input. Some input is more accurate than others, and some are just wrong. I am striving to provide the highest quality data I can.

That's not what the Owners Manual says:

"Warning: Do not use Traffic-Aware Cruise Control on city streets or on roads where traffic conditions are constantly changing."

There is always a segment of the population that will disregard instructions. I would rather the enthusiasts at TMC push every button and find every use case before a Model 3 owner who comes from Honda/Toyota tries the same. :D
 
Dropped my X off to get the windshield tinted and have a few more observations.

- I can confirm TACC on AP2 Hardware ignores immobile objects. If you are driving down a highway/city street where a car is stopped in front of you, you will hit it at full speed. This was known to some AP1 drivers but news to me.

Tesla has to do something about this. I know there will be a threads one day "TACC caused me to rear end a stopped vehicle" or "Family sues Tesla after car rear ended and two children died".

- Autosteer will drop your cruise control speed down to 35. I had TACC set to 70. Traffic slowed down to 25. I turned on autosteer and saw my new cruise control speed was 35. I accelerated back up to past 35 where auto steer turns off. Not a problem, just an observation.

- TACC was set to 65 and I took a ramp rated for 50 No adjustments in speed. I took a second ramp where speed did "appear to drop". There was however a vehicle in front and the curve was much tighter. This needs more data to confirm/deny TACC behavior/speed on ramps.
 
Just want to report on how AP1 behaves regarding slowing down for curves. It definitely slows down for curves even when only TACC is engaged. It does not slow all the way down to the speed posted on yellow warning signs but will typically be about 10 MPH above that. It even slows for curves without a warning sign. I think it uses the map to know the curvature of the road and hence the appropriate speed. It works pretty well, but isn't perfect. I pay close attention and do slow manually sometimes.
 
View attachment 208791

Edit: So, I saw no mention of time, car length, feet, meters, and it should adjust to speed. Call it Elon distance.

You obviously haven't taken any higher math or science classes. Distance over time is speed. Any sentence saying that distance will vary based on your speed is implying a time element because you can't have distance without time.

You can complain that you want it stated in a different format but the statement on that screenshot gives enough information to know that it is a time based setting.
701b935ef4072b0f79c429a0d461a6cce437f1c2.gif

The triangles will help you remember these 3 rules:
  • Distance = Speed x Time
  • Time = Distance/Speed
  • Speed= Distance/Time
 
Last edited:
Even on 'early' I have not had seen any bonafide activation's. The 3 false positive activation's had no vehicle in front of me at all.

It would be good to know more specific setting -> behaviors rather than

Follow Distance 1-7. What does 1-7 mean? I had to estimate 7 being 1.3-1.5 seconds using my Volt as the measuring stick.

What does early and medium correspond to for seconds before collision?

Sensor range of ultrasonics is fixed so if those are used for collision warning / AEB then the time will vary with speed.

If the warning is from radar the range is also fixed (practically as accuracy drops off after a certain range) but it is a longer range sensor than the ulrasonics. I would still expect variance in time to impact.

If the warning is from camera then processing power and focal length might be the limiting factor. I'd expect this one would not vary in time as much as the other two but who knows maybe the radar will pick up something before the camera.

End result you can't put a fixed time label on a setting when the cars speed varies and the sensors have range limitations.
 
Last edited:
Perhaps Tesla should take a tip from the 82nd Airborne. The guys stuffing the parachutes are the first ones out the door of the plane with a random parachute packed by someone else. If the Tesla guys writing the code were all given a Tesla and forced to eat their own dog food, things would probably get better quickly.

See post #55 in this very thread. 8.0 (2.50.185) caution using TACC/Autosteer features "Eating their own dogfood" is done at Tesla.

I think that will intensify when all the SpaceX, Tesla, Solarcity people get their Model 3 before the rest of us.
 
  • Funny
Reactions: lunitiks
Sensor range of ultrasonics is fixed so if those are used for collision warning / AEB then the time will vary with speed.

If the warning is from radar the range is also fixed (practically as accuracy drops off after a certain range) but it is a longer range sensor than the ulrasonics. I would still expect variance in time to impact.

If the warning is from camera then processing power and focal length might be the limiting factor. I'd expect this one would not vary in time as much as the other two but who knows maybe the radar will pick up something before the camera.

End result you can't put a fixed time label on a setting when the cars speed varies and the sensors have range limitations.

What would help is visual indicators of what Tesla Vision 'sees' without the need for audible warnings that are outside of the warning threshold (Early/Medium/Late). For example, I think the parking sensors go from White -> Orange -> Red -> STOP. It would give the drivers more confidence that the features work or can work when the time comes. No one in their right mind would repeatedly try to trigger FCW or AEB.

On a side note, my Volt has FCW and AEB and I put 0 trust in it. I get false positives on FCW and no triggers on FCW where I think it 'should' trigger. I don't have access to many vehicles so I don't which manufacture does this type of feature best.
 
I can confirm TACC on AP2 Hardware ignores immobile objects. If you are driving down a highway/city street where a car is stopped in front of you, you will hit it at full speed.
This is really concerning, am assuming it is not true if the lead vehicle that AP is tracking comes to a halt on its own.

My overall read of this thread is that Tesla really hacked the video they put on their website showing off the AP2 hardware's capabilities and had all (at least me) excited about buying the car with AP2 hardware. Despite a supercomputer on board and all the multi million miles of AP1 driving data available, it seems to be crawling at the implementation of basic self driving. This will not go well with Tesla if it doesn't quickly get back on its feet.
 
  • Disagree
Reactions: HumanGenome