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

Follow-the-leader discussion out of Market Action

mongo

Well-Known Member
May 3, 2017
13,296
40,663
Michigan
Confirms what we already know except that the system did not delivery any warning at all (not even a late one). That last one is concerning and an obvious target of improvement.

Much discussion over here:
Model X Crash on US-101 (Mountain View, CA)

Sounds like it was similar to the other 'following a vehicle that changes lanes and there is a stationary object ahead' collisions
Ideally, the new EAP vision NN will have some form of in-lane stopped object detection.
 
  • Informative
Reactions: neroden

bdy0627

Active Member
May 19, 2015
3,505
12,087
Appleton, WI
Much discussion over here:
Model X Crash on US-101 (Mountain View, CA)

Sounds like it was similar to the other 'following a vehicle that changes lanes and there is a stationary object ahead' collisions
Ideally, the new EAP vision NN will have some form of in-lane stopped object detection.
Or at the very least a warning to the driver in these situations where the AP loses the lane and tracks the lead car, since that can be a risky situation when the lead car changes lanes. Simply warning the driver that the car is no longer identifying the lane and instead tracking the lead car would hopefully get the driver to be particularly vigilant in that situation. As it is now, that process happens quietly.
 

lklundin

Active Member
Oct 10, 2014
2,973
20,335
Bavaria
  • Like
Reactions: neroden

Esme Es Mejor

Member
Dec 18, 2016
837
5,005
Bay Area
Or at the very least a warning to the driver in these situations where the AP loses the lane and tracks the lead car, since that can be a risky situation when the lead car changes lanes. Simply warning the driver that the car is no longer identifying the lane and instead tracking the lead car would hopefully get the driver to be particularly vigilant in that situation. As it is now, that process happens quietly.

Yeah, it seems like a warning chime to alert the driver would be very useful in that situation, indicating : “I’ve stopped following the lead car— please make sure I’m in the correct lane.” Some drivers would continue being inattentive anyway, but it would help focus some drivers.
 
  • Like
Reactions: Sudre

neroden

Model S Owner and Frustrated Tesla Fan
Apr 25, 2011
14,676
62,629
Ithaca, NY, USA
Based on that NTSB description, veteran users of AP will understand exactly what happened:

1. Model X was following a lead car.
2. That lead car was in the "non-travel lane", finally realized their mistake, and moved over at 7 seconds prior to the crash. That's why the steering begin to shift left--it was trying to follow the lead car, but saw the lane markings and then stopped following the lead car.
3. Model X then sped up to driver's set speed as it wasn't following a lead car anymore.
4. When a lead car changes lanes, AP can take a little extra time for the radar to lock on a new target (and stationary targets will be disregarded). That's a time where the driver needs to be paying extra-close attention. Same issue with the firetruck. AP following a car, car changes lanes, AP car runs into firetruck.

Recommendations:

1. Driver must stay engaged.
2. Tesla should work on rapid acquisition of stationary objects in radar's path of travel.
3. CalTran needs to update their road markings and improve signage/safety in that area.

Note that this scenario (lead car changes lanes to reveal a hazard) is explicitly covered in the manual. While Tesla should work on it, I wouldn't be too concerned about restrictive action. He who follows not the manual may experience negative results.

Case closed.

This is the second time I've closed an NTSB Autopilot investigation early for them. I think they need to start paying me.

Interesting. If this "lead car" business is correct, then I fundamentally disagree with Tesla's entire algorithm. You should never "follow a lead car" -- only bad drivers do that. You stay in your own lane independently, regardless of what the other idiots are doing, and then you adjust your driving to keep extra distance from any cars near you (starting with any car in front, followed by any car on either side, then any car in the diagonal forward positions, then any car directly behind you, then any car diagonally behind you).

This "lead car" algorithm is an error. I don't know whose idea it was but it's based on bad driving. I have seen humans drive that way, and they are *wrong*.

This is what I don't like about the extremely incompetent automated driving programs so far; they seem to be designed by people who are bad drivers, to replicate bad driving behavior. It's really not acceptable.
 

neroden

Model S Owner and Frustrated Tesla Fan
Apr 25, 2011
14,676
62,629
Ithaca, NY, USA
This sounds very likely to me from the data we have seen. One thing Tesla could do that would be very helpful in this situation where the lead car moves over and the car tries to find it's bearings again is to sound an alert any time it is having trouble finding the lane and shifts to following a lead car.
Agreed. I'd also give the driver the option to turn off the "following a lead car" option, as it is generally wrong. Perhaps always wrong.

I have driven through really bad snowy weather where the cars in front of me were doing crazy stuff and mostly ending up in the ditch, either due to bad traction or not being able to see the road. I ignored them apart from not running into them and took my cues for where the lane was from other things, which is correct.

This happens pretty seamlessly as it is now so that the driver only notices if he looks at the display and sees the car ahead colored in blue.
Yuck. This is an actual design error and one which creates unsafe situations.

That is a situation where the driver needs to be particularly attentive because the Autopilot is having difficulty identifying the lane. It's one of the worst times for the driver to zone out. If the car would sound an alert for the driver, it could help make sure the driver is focusing attention on the road in this risky situation. That would be an easy, inexpensive implementation that could save lives when this situation occurs.
 

mulder1231

Active Member
Jan 1, 2012
1,814
10,164
San Diego
Interesting. If this "lead car" business is correct, then I fundamentally disagree with Tesla's entire algorithm. You should never "follow a lead car" -- only bad drivers do that. You stay in your own lane independently, regardless of what the other idiots are doing, and then you adjust your driving to keep extra distance from any cars near you (starting with any car in front, followed by any car on either side, then any car in the diagonal forward positions, then any car directly behind you, then any car diagonally behind you).

This "lead car" algorithm is an error. I don't know whose idea it was but it's based on bad driving. I have seen humans drive that way, and they are *wrong*.

This is what I don't like about the extremely incompetent automated driving programs so far; they seem to be designed by people who are bad drivers, to replicate bad driving behavior. It's really not acceptable.

Follow the lead car is only happening in AP if no lane detected.
 

neroden

Model S Owner and Frustrated Tesla Fan
Apr 25, 2011
14,676
62,629
Ithaca, NY, USA
Follow the lead car is only happening in AP if no lane detected.
This algorithm should probably just be removed. :-( It's not a safe backup algorithm. I've seen too many incompetent drivers on the road; I remember more than one occasion, with no lane markings visible, where the car directly in front of me ran into the opposing lane (on an undivided highway), or into the ditch, or whatever. I didn't follow them.

If they're using this as a long-term test for convoying, they should only allow it to follow other Teslas running on autopilot, not normal cars. ;-)

It's good to know what's causing the problem. It's a fundamental design error, but a simple enough one to change; once someone in charge recognizes that "follow the car in front of you" is never a good rule, we should stop having these crashes.
 

bdy0627

Active Member
May 19, 2015
3,505
12,087
Appleton, WI
This algorithm should probably just be removed. :-( It's not a safe backup algorithm. I've seen too many incompetent drivers on the road; I remember more than one occasion, with no lane markings visible, where the car directly in front of me ran into the opposing lane (on an undivided highway), or into the ditch, or whatever. I didn't follow them.

If they're using this as a long-term test for convoying, they should only allow it to follow other Teslas running on autopilot, not normal cars. ;-)
I agree. I would rather have AP simply beep at me to take over when it can no longer detect the lane. I do not like simply following the lead car and assuming he is driving well. At the very least, I would like a warning when it switches over so I can take over immediately if I want.
 

ruralectric

Member
Mar 27, 2016
228
1,414
Eastern Washington
Interesting. If this "lead car" business is correct, then I fundamentally disagree with Tesla's entire algorithm. You should never "follow a lead car" -- only bad drivers do that. You stay in your own lane independently, regardless of what the other idiots are doing, and then you adjust your driving to keep extra distance from any cars near you (starting with any car in front, followed by any car on either side, then any car in the diagonal forward positions, then any car directly behind you, then any car diagonally behind you).

This "lead car" algorithm is an error. I don't know whose idea it was but it's based on bad driving. I have seen humans drive that way, and they are *wrong*.

This is what I don't like about the extremely incompetent automated driving programs so far; they seem to be designed by people who are bad drivers, to replicate bad driving behavior. It's really not acceptable.
I've wondered about this myself. It does seem like a vulnerability. (Owe look, I'm being followed by a Tesla! What happens if I engage in a bit of lane weaving?) Serious question: Does anyone know what happens in this case?
 
  • Informative
Reactions: neroden

mongo

Well-Known Member
May 3, 2017
13,296
40,663
Michigan
This algorithm should probably just be removed. :-( It's not a safe backup algorithm. I've seen too many incompetent drivers on the road; I remember more than one occasion, with no lane markings visible, where the car directly in front of me ran into the opposing lane (on an undivided highway), or into the ditch, or whatever. I didn't follow them.

If they're using this as a long-term test for convoying, they should only allow it to follow other Teslas running on autopilot, not normal cars. ;-)

It's good to know what's causing the problem. It's a fundamental design error, but a simple enough one to change; once someone in charge recognizes that "follow the car in front of you" is never a good rule, we should stop having these crashes.

I hear it is useful for large unlined intersections (assuming the lead car knows what they are doing).
Ob xkcd
https://xkcd.com/537/
 
  • Love
  • Funny
Reactions: neroden and MP3Mike

EinSV

Active Member
Feb 6, 2016
4,331
21,540
NorCal
I've wondered about this myself. It does seem like a vulnerability. (Owe look, I'm being followed by a Tesla! What happens if I engage in a bit of lane weaving?) Serious question: Does anyone know what happens in this case?

As a practical matter, I can't recall the switch from lane following to car following and back ever being a problem for me -- it's pretty smooth.

And anytime AP does anything hinky, which is fairly rare, you take over. Not a big deal IMO.

FSD is a different story, but we're not there yet.

Looking forward to some statistics from Tesla next month. It is easy to get caught up with perceived vulnerabilities and lose sight of the big picture, which can only be shown by data.
 

defc0n

Active Member
Sep 30, 2016
1,318
3,914
Indiana
This algorithm should probably just be removed. :-( It's not a safe backup algorithm. I've seen too many incompetent drivers on the road; I remember more than one occasion, with no lane markings visible, where the car directly in front of me ran into the opposing lane (on an undivided highway), or into the ditch, or whatever. I didn't follow them.

If they're using this as a long-term test for convoying, they should only allow it to follow other Teslas running on autopilot, not normal cars. ;-)

It's good to know what's causing the problem. It's a fundamental design error, but a simple enough one to change; once someone in charge recognizes that "follow the car in front of you" is never a good rule, we should stop having these crashes.

It’s all a game of probabilities. Statistically, if no lane marker is detected, what is your next best move? What would you have it do otherwise? Should autopilot give up? Get on the brakes, go in a direct straight line even if the lane might be turning?

“I dunno for sure exactly I’m doing so I’m just gonna follow that guy”

That guy might **** up, but statistically he’s your best bet.
 

2virgule5

Supporting Member
Supporting Member
Dec 6, 2013
667
1,912
Union county NJ
This algorithm should probably just be removed. :-( It's not a safe backup algorithm. I've seen too many incompetent drivers on the road; I remember more than one occasion, with no lane markings visible, where the car directly in front of me ran into the opposing lane (on an undivided highway), or into the ditch, or whatever. I didn't follow them.

If they're using this as a long-term test for convoying, they should only allow it to follow other Teslas running on autopilot, not normal cars. ;-)

It's good to know what's causing the problem. It's a fundamental design error, but a simple enough one to change; once someone in charge recognizes that "follow the car in front of you" is never a good rule, we should stop having these crashes.
In a situation such as the one you are describing AP would NOT follow the car drifting rapidly right or left. It is solely used for few seconds in the absence of clear lane markings - if lanes are not recaptured by the system it would disengage (« take over immediately »).
Also, in my few thousands miles of AP 2.0 I never saw the ‘blue’ car icon - I’m actually not sure that this algorythm that was introduced to compensate for the shortcomings of AP1 is still active.
 

Todd Burch

Voltage makes me tingle.
Nov 3, 2009
8,147
32,480
Smithfield, VA
Neroden, I think you’re oversimplifying the system a bit.

The lane following thing is not black and white. It’s a weighted uncertainty.

If you’re going less than a few mph on AP and following a car, the car will tend to move to be directly behind the car in front. This seems to be for a few reasons:

1. The lane lines are harder for the cameras to see (if at all) because you’re close to the lead car. The lead car blocks a good view of the lane lines.

2. If lead cars are going around an obstruction on the ground (like a pothole), you want to go around it too.

Once the car speeds up a bit (about 10-15 mph) it will tend to follow lane lines more because following distance allows AP to see the lines better.

At higher speed driving, there is a weighted mix going on. Following distance is larger so the cameras can make out more of the road path from the lane lines. If lane lines are detected with high certainty, the lead car is not used much for directional guidance.

The less certainty the system has that it knows the lane lines, the more weight it puts on following a lead car. This is especially noticeable when going through unmarked intersections: if the lead car changes lanes while you’re in an unmarked intersection, your car will start to follow it and change lanes also (unless there are cars adjacent to you within radar/ultrasonic range which provide directional guidance). So that’s another scenario to be careful about.

AP depicts this on the instrument cluster: items not providing guidance are grey, things providing guidance are white, and the thing providing primary guidance (vehicle or lane lines) is blue.

If a lead car is blue, it means you’re following it. If the lane lines are blue, the car is following the lane lines.

This weighted uncertainty is important also when you go through an area with weakly painted lines. In that case, certainty for lane detection is low, but a lead car will probably provide better guidance.

In the Moutain View case, uncertainty on the lane lines was high enough that the X started to follow the lead car until enough of one of the lane lines was visible to the camera to provide higher weighting to the lane markings, at which point the lanes took over guidance and the car straightened in the lane.

Anyway, with the available computing power of AP1, I think they’re doing about as much as they can do.
 

neroden

Model S Owner and Frustrated Tesla Fan
Apr 25, 2011
14,676
62,629
Ithaca, NY, USA
It’s all a game of probabilities. Statistically, if no lane marker is detected, what is your next best move? What would you have it do otherwise? Should autopilot give up? Get on the brakes, go in a direct straight line even if the lane might be turning?
If I really can't detect the lane, I hit the brakes.

Normally before doing that I look for several backup sources of lane position: the markers at the side of the road are my first go-to choice, followed by the road surface texture/where the snow is pile, the mailboxes, and finally the marks on the road/snow made by other cars, if I can't find anything else.

On a multi-lane road, I first identify where the *road* is, and if I can do that, I then make a logical guess regarding the lane layout of the road based on general informed knowledge.

“I dunno for sure exactly I’m doing so I’m just gonna follow that guy”

That guy might **** up, but statistically he’s your best bet.
No he isn't.

This is why we need to have these things being trained by competent drivers, not by people who use bad algorithms.
 
  • Like
Reactions: Sudre

neroden

Model S Owner and Frustrated Tesla Fan
Apr 25, 2011
14,676
62,629
Ithaca, NY, USA
Neroden, I think you’re oversimplifying the system a bit.

The lane following thing is not black and white. It’s a weighted uncertainty.

If you’re going less than a few mph on AP and following a car, the car will tend to move to be directly behind the car in front. This seems to be for a few reasons:

1. The lane lines are harder for the cameras to see (if at all) because you’re close to the lead car. The lead car blocks a good view of the lane lines.
Also known as "following too close"

2. If lead cars are going around an obstruction on the ground (like a pothole), you want to go around it too.
Actually you want to be far enough back to detect the pothole on your own even if the car in front of you hits it directly. Particularly if it does.

Once the car speeds up a bit (about 10-15 mph) it will tend to follow lane lines more because following distance allows AP to see the lines better.

At higher speed driving, there is a weighted mix going on. Following distance is larger so the cameras can make out more of the road path from the lane lines. If lane lines are detected with high certainty, the lead car is not used much for directional guidance.

The less certainty the system has that it knows the lane lines, the more weight it puts on following a lead car. This is especially noticeable when going through unmarked intersections: if the lead car changes lanes while you’re in an unmarked intersection, your car will start to follow it and change lanes also (unless there are cars adjacent to you within radar/ultrasonic range which provide directional guidance). So that’s another scenario to be careful about.

AP depicts this on the instrument cluster: items not providing guidance are grey, things providing guidance are white, and the thing providing primary guidance (vehicle or lane lines) is blue.

If a lead car is blue, it means you’re following it. If the lane lines are blue, the car is following the lane lines.

This weighted uncertainty is important also when you go through an area with weakly painted lines. In that case, certainty for lane detection is low, but a lead car will probably provide better guidance.

In the Moutain View case, uncertainty on the lane lines was high enough that the X started to follow the lead car until enough of one of the lane lines was visible to the camera to provide higher weighting to the lane markings, at which point the lanes took over guidance and the car straightened in the lane.

Anyway, with the available computing power of AP1, I think they’re doing about as much as they can do.

If that's the best they can do, it's not ready to use. And I don't think it's the best they can do; this isn't a computing power problem, this is a fundamental design error. This is *bad driving*. If I saw a human following this algorithm (and to be fair, I have), I would tell the human to cut it out.

I'm glad the instrument cluster shows what AP is "thinking", but it's doing a crap job, because it's been programmed by incompetent drivers.

Primary guidance when driving should be, first, the *detected stationary objects* which you are avoiding, followed by the *calcuated blind spots* which you are *also* avoiding (because they might contain things you can't see). This is how I learned to drive. After that you start dealing with such things as moving objects which you are avoiding, and finally the route you're actually trying to take.

I'm glad that there are so many people here who seem to have a good understanding of how Tesla's Autosteer is currently working. As currently done, it's really unacceptable and should be scrapped. Thankfully, it actually seems like a simple problem to get a decent system working -- the error is that they've told it to do the wrong thing, so tell it to do the right thing and you can probably get it working.
 

Products we're discussing on TMC...

About Us

Formed in 2006, Tesla Motors Club (TMC) was the first independent online Tesla community. Today it remains the largest and most dynamic community of Tesla enthusiasts. Learn more.

Do you value your experience at TMC? Consider becoming a Supporting Member of Tesla Motors Club. As a thank you for your contribution, you'll get nearly no ads in the Community and Groups sections. Additional perks are available depending on the level of contribution. Please visit the Account Upgrades page for more details.


SUPPORT TMC
Top