3 day old import P85D crashed while using TACC

[AnxietyRanger]

It recognized the stationary object. But tacc can't stop for every stationary object dead ahead because usually people drive around them. That is the issue. If the car that was being followed drove around the stopped car the tesla car could have done the same in which case slowing down and stopping would not have been desirable

Maybe I missed it, but how can you know it recognized the stationary object?

In my experience ACC implementations usually err on the side of caution, at the very least not accelerate when something seems to be ahead, maybe slow down or continue a slower speed for a while.

- - - Updated - - -

This! I do the same and as soon as I let go off the pedal, it resumes what it was doing. Nanny it a little bit and the experience is great.

I also like that the car errs on the side of not stopping if its not sure what is in front. I can think of a few scenarios in my daily drive where the car could abruptly apply the brakes and possibly get me into a rear-end accident with drivers behind me. The emergency brakes algorithm of minimizing impact only at the last possible second is good enough for me. At this point, I am not expecting the car to be very autonomous.. just a friend who helps out here and there.

False positivies as in emergency braking certainly can be dangerous in their own right, but that isn't the only alternative. Another would be ACC erring on the side of slowing down manageably, unless driver pushes the pedal to override the cautionary measure. In my experience, this is how the different Germans I've driven do it.

An interesting aside: Audi for example watches blinkers on the car in front to determine if they may be leaving the road and adjust ACC accordingly.

- - - Updated - - -

[Addressing the part in bold above.] Not necessarily. There are ambiguous situations due to the fact that the human driver is still controlling the steering. Consider the case of a fork in the road (or lane split or exit ramp, etc.), which is apparently the situation the OP encountered. Maybe there is a straight ahead branch of the fork with stopped traffic and a slight right branch with no traffic. You have TACC enabled and are following a vehicle that TACC has locked onto. That vehicle bears right at the fork. The car has no way of knowing whether you plan to follow the moving car or take the other branch with the stopped traffic. If it's the former, slamming on the brakes would be a bad idea. If it's the latter, the car may not have enough distance to avoid a collision, even if Automatic Emergency Braking kicks in (which it should at some point when it detects the collision is unavoidable--unless there isn't even enough time for that). There are any number of scenarios where AEB will not be able to avoid a collision (different fork angles, speeds, following distances, road/tire conditions, etc.).

Smarter features incorporating more sensor inputs will be added over time. More/better sensors will be added to get even smarter functionality. Mobileye EyeQ and NVIDIA Drive PX have some impressive demos. The driver will still be responsible for the safe operation of the vehicle and understanding how the driver assistance features work.

Just some notes:

Germans use map data to determine the situation, for example behaving differently near motorway exists and crossroads. They do tend to err on the side of slowing down, but may delay slowing down when they determine a car is leaving the highway in front (blinkers on and an exit on map), for example. I'm not sure if Tesla is this smart.

You are right Tesla's sensor suite currently is quite limited. High-end Germans usually have two or more radars in both ends of the car (plus the camera and ultrasonics), Tesla only has one radar in front (plus the camera and ultrasonics).

- - - Updated - - -

Do you mean that the TACC display goes blue at some point when you roll up on a stopped car that doesn't move at all? I've never seen that happen, but perhaps I always bail out and press the brake before that. I'm going to find out how much risk I'm willing to experience in the name of science on the way home this afternoon.

The new Audi Q7 uses camera for full-stop braking (it is the first car in the world to do it, previously it was done with radars/ultrasonics). It has the same EyeQ3 as the auto-pilot Model S, so the camera technology shouldn't in itself be an issue.

 

[CHG-ON]

I am so sorry to hear about your accident and I am relieved that you are all OK. I would be devastated.

I use TACC in all sorts of situations and find that it generally works quite well. But I certainly don't trust it. I always keep hyper vigilant about everything going on around me. I actually leave my foot on the go pedal so that it is in the optimal position to react. I have had situations where it has not slow as quickly as I would have liked in a few situations and have intervened without incident. I actually had a somewhat opposite problem in slow-and-go traffic on the highway. We were rolling along under TACC at perhaps 5-10MPH, the car in front of me started to slowly accelerate as traffic lightened up and my car thought there was an impending collision and slammed the brakes on and sounded the alarm. Fortunately, I was paying full attention and was able to hit the accelerator quickly enough to avoid being rear ended by the car behind me. I reported to Tesla immediately.

What really scares me is when we have autonomous accel, braking AND steering. How many people will rest with the false sense of security that the car knows what it is doing and will not make a mistake? It's not going to be pretty. Americans already don't pay attention to what they are doing when driving without these aids. I think this tech will need another decade before we could say that it is 100% safer than human controlled driving. This terrible incident is a perfect example of how flawed this tech continues to be in these early days. Google seems to be on the right track by testing, testing, testing before releasing anything. I think the auto manufacturers are jumping the gun.

Thank goodness that you are alright. Even though the car cost a fortune and it is a heartbreaker, you and your family are OK. I hope that you will be able to sort out the loss to your satisfaction.

 

[stopcrazypp]

Now now, don't jump to conclusions. I am all ears for further clarification and acknowledge such when offered, unlike some people.

The scenario you describe makes sense. The camera could still have seen the car in front, though, this was not a 2005 one-radar ACC. But I agree, the scenario you paint makes sense and recognizing stationary objects is still limited on the average ACC, although not at all impossible.

After seeing bhzmark's quote describing the OP's situation (car it was following moved out of lane revealing a stopped vehicle), it's pretty clear it's exactly the same situation where both the Mercedes and BMW manual says ACC will *not* brake for (and the wording is clear, it says *does not* explicitly; it does not say *may not*).

The issue is not necessarily detection, it's the logic of the system. It assumes you will follow the car that left the lane rather than stay in the lane with the stationary object. By the time it realizes it must brake to avoid impact, it's the auto-braking system that activates (not ACC) and that one is only designed to minimize impact, but not necessarily prevent impact.

Yes, by the time autopilot is out, it must have the logic to figure out this situation (the equipment must recognize the stopped vehicle as a stopped vehicle and be able to make the assumption the driver will stay in their lane unless the turn signal is pressed, since there is auto-steering).

 

[AnxietyRanger]

After seeing bhzmark's quote describing the OP's situation (car it was following moved out of lane revealing a stopped vehicle), it's pretty clear it's exactly the same situation where both the Mercedes and BMW manual says ACC will *not* brake for (and the wording is clear, it says *does not* explicitly; it does not say *may not*).

The issue is not necessarily detection, it's the logic of the system. It assumes you will follow the car that left the lane rather than stay in the lane with the stationary object. By the time it realizes it must brake to avoid impact, it's the auto-braking system that activates (not ACC) and that one is only designed to minimize impact, but not necessarily prevent impact.

Yes, by the time autopilot is out, it must have the logic to figure out this situation (the equipment must recognize the stopped vehicle as a stopped vehicle and be able to make the assumption the driver will stay in their lane unless the turn signal is pressed, since there is auto-steering).

I agree, it is true that the scenario of a stationary car appearing behind a moving car is a scenario radar-based ACC hasn't been able to tackle or at least one many have excluded in manuals. Not even in the case of Distronic Plus that has multiple radars (Model S has only one) is that a clear-cut case. That said, Model S does have a front-facing camera with EyeQ3 chip that definitely technically is capable of this, but it is possible it is "dumb" so far and doesn't try. There is no reason it couldn't: new Audi Q7 does stopping based on camera with EyeQ3 alone, so it can be done. EyeQ3 can recognize pretty much any shape in traffic and label it as a car or a pothole or whatever. The hardware is there.

All this said, what remains intriguing to me in this case is not necessarily the fact that the Tesla hit the stationary object, which as you say is understandable given the nature of radar-based ACC, but the suggestion it accelerated after the car in front changed lanes. My experience with the late-generation Germans suggest they are quite cautious to accelerate in these kinds of situations, waiting painfully long before "hitting it" again. Was the driver too preoccupied or did the car indeed react (accelerate) in an overly aggressive manner after the lane change in front... or perhaps was it just about the car in front changing lanest at the last minute revealing another car behind...?

Alas, we hardly know enough, either of the Tesla vs. existing ACC accuracy, or of the case in question. My own experience with several German ACC systems, though, is that they have saved my butt more often than they have threatened it - even when the limits of the system on paper were exceeded. My gut guesses a 2015 S Series - with its vast sensors and more mature software - would have stopped in time where 2015 Model S failed to do so, but I hardly know enough of this particular case to be sure, which goes for most everyone else in the thread too. (Not sure if U.S. S Series has the same sensors as an European one, but since the Tesla was EU version too, it seems a worthy comparison.)

 

[NSX1992]

I use TACC always on the freeway and trust it as it always worked . However a couple of weeks ago I accelerated to 80 mph from my on ramp to the #1 lane in wide open traffic. However the cars in front all of a sudden had stopped and I was relying on TACC to stop me. When I finally realized that it wasn't happening I slammed on the brakes. That was the first time I felt the ABS brakes working including skidding and smelled rubber. I and the cars in front and back were all able to stop in time but now I am prepared to take control sooner.

 

[Andyw2100]

I use TACC always on the freeway and trust it as it always worked . However a couple of weeks ago I accelerated to 80 mph from my on ramp to the #1 lane in wide open traffic. However the cars in front all of a sudden had stopped and I was relying on TACC to stop me. When I finally realized that it wasn't happening I slammed on the brakes. That was the first time I felt the ABS brakes working including skidding and smelled rubber. I and the cars in front and back were all able to stop in time but now I am prepared to take control sooner.

You say "wide open traffic." Had TACC acquired a target yet?

 

[AnxietyRanger]

The issue is not necessarily detection, it's the logic of the system. It assumes you will follow the car that left the lane rather than stay in the lane with the stationary object. By the time it realizes it must brake to avoid impact, it's the auto-braking system that activates (not ACC) and that one is only designed to minimize impact, but not necessarily prevent impact.

Re-reading our messages of yesterday, I must say I don't agree with this particular part.

As said, I agree what happened may have been this, and for radar-only ACC this is a challenging situation:

However, what I don't agree is that the ACC would assume you want to follow. The ACC will know your car is not pointed at the car that is changing lanes, the car will assume you'll go forward as your car is pointing there. It accelerates because it sees the obstacle in front (that was tracked) moving away from in front. So I don't think the ACC assumes you will follow the car that left the lane if you keep pointing forwards. Tesla's single central radar also has a fairly narrow field of view, the car in front probably left it quite quickly - many Germans use several radars to see wider, but Model S has only one.

So, while I agree on the likely scenario, I don't think the average ACC assumes you want to follow the car that left the lane. I think ACC assumes you go where your steering wheel points.

All this said, I've found German ACC implementations to be fairly cautious after a car in front has left the lane - to the point that I often nurse them with the pedal to go faster. I have a hard time seeing how this accident would have happened especially with a camera-equipped ACC from Audi or Mercedes Benz.

 

[bhzmark]

Stopcrazy has it exactly rt. In your very nice illustration the ACC can't distinguish that between the road simply curving to the right and the car in front being on the shoulder of the road. We don't want our cars to slow down for cars parked on the side of a curved road.

 

[AnxietyRanger]

Stopcrazy has it exactly rt. In your very nice illustration the ACC can't distinguish that between the road simply curving to the right and the car in front being on the shoulder of the road. We don't want our cars to slow down for cars parked on the side of a curved road.

One point, though: The ACC knows where your steering wheel is pointed at. The story in this case is, it even accelerated because it sensed the car in front moving away - that does not seem consistent with "follow" either. I think the car thought, correctly, you'll want to continue forwards. Where it failed, though, is seeing the stopped car in front (normal for a mere radar, although the forward-facing camera can see a stationary car).

The thing that troubles me with this story is:

a) Doesn't Tesla use the forward camera to help out the ACC, maybe not yet - the camera with EyeQ3 chip could recognize the parked car
b) Does Model S accelerate too aggressively after a car in front changes lanes (my experience from Germans being they are very cautious in this scenario)

But of course, it is also possible the driver in question trusted the system way too much and expected the car to stop for the parked car that emerged, was pre-occupied and failed to react in time or the car in front changed lanes very close to the stopped car in front.

 

[bhzmark]

One point, though: The ACC knows where your steering wheel is pointed at. The story in this case is, it even accelerated because it sensed the car in front moving away - that does not seem consistent with "follow" either. I think the car thought, correctly, you'll want to continue forwards. Where it failed, though, is seeing the stopped car in front (normal for a mere radar, although the forward-facing camera can see a stationary car).

The thing that troubles me with this story is:

a) Doesn't Tesla use the forward camera to help out the ACC, maybe not yet - the camera with EyeQ3 chip could recognize the parked car
b) Does Model S accelerate too aggressively after a car in front changes lanes (my experience from Germans being they are very cautious in this scenario)

But of course, it is also possible the driver in question trusted the system way too much and expected the car to stop for the parked car that emerged, was pre-occupied and failed to react in time or the car in front changed lanes very close to the stopped car in front.

If the driver was following the car turning to the right, he would still be aiming roughly for the stopped car as well -- until he finally did turn. That isn't a sufficient indication.

Even if the camera input can be further dialed in to improve the logic and help TACC better guess which way you intend to go (don't stop for parked cars on the shoulder, but do stop for parked cars in current lane of travel, even if previously (currently?) tracked car went around them -- but what if the lane simply splits?). While steering control is with the driver and throttle is with the autopilot they will not always match up with the driver intention which is inherently unknowable -- at least until the car can do the steering also and know for sure which way the car is going.

 

[AnxietyRanger]

In my comments I assumed the OPs friends car was driving straight, was to continue straight, and the car in front diverged significantly when it turned away (to reveal a stopped car in front).

If it was a bend situation then things of course can be different. I've noticed that German ACC implementations may delay acceleration in bends exactly for this kind of reason, by the way.

Check out EyeQ3 capability videos sometime. It is amazing what cameras like that (or the future DRIVE PX which does it in color) can gather from a driving scene. A lot. Tesla probably doesn't use the camera much yet.

 

[wk057]

I've skimmed through the comments in this thread and I just feel that some main points need to be brought back to the forefront.

• TACC is not autonomous driving. It's cruise control that will attempt to slow for a slower car in front when conditions allow. Conditions don't always allow, as is made abundantly in the manual including specifics to this situation.
• TACC is not designed for use on surface streets. People using it in those areas and getting into accidents or near accidents as a result of their inattentiveness deserve to be in accidents, IMO.
• This accident is 100% the fault of the driver. 100%. Not 99%. 100%. Had the driver paid attention as required in the situation and reacted like any reasonable driver would then there wouldn't have been a problem.

Driver assistance, not driver replacement. Get a grip folks. The OP screwed up, end of story. An unfortunate screw up, but one none the less.

 

[dsm363]

I've skimmed through the comments in this thread and I just feel that some main points need to be brought back to the forefront.

• TACC is not autonomous driving. It's cruise control that will attempt to slow for a slower car in front when conditions allow. Conditions don't always allow, as is made abundantly in the manual including specifics to this situation.
• TACC is not designed for use on surface streets. People using it in those areas and getting into accidents or near accidents as a result of their inattentiveness deserve to be in accidents, IMO.
• This accident is 100% the fault of the driver. 100%. Not 99%. 100%. Had the driver paid attention as required in the situation and reacted like any reasonable driver would then there wouldn't have been a problem.

Driver assistance, not driver replacement. Get a grip folks. The OP screwed up, end of story. An unfortunate screw up, but one none the less.

Yep. There really isn't much more to say. Maybe Tesla should reinforce this on delivery but the OP imported the car into an unsupported country so probably didn't undergo the normal delivery process.

 

[ecarfan]

@wk057, I agree with your points, but would add one more:

This unfortunate ate accident highlights another reason why it is risky to import a Tesla into a country that Tesla does not sell in. The owner did not take delivery from Tesla and therefore did not have a Tesla employee provide them with an orientation to the car. Surely that orientation would include information about the proper use of the TACC, since that feature is new to anyone buying a Tesla these days, and if the feature is not used correctly accidents can result. It's not like learning how to use the climate control or the navigation screen: TACC is a safety-critical feature that is unlike anything on 99% of other car models currently sold.

A thorough reading of the manual on an advanced car like the Model S is essential before operating it for the first time. I read every page of the manual before I got my S. Of course I could access the manual through the "My Tesla" account before taking delivery. Since the OP did not buy his car directly from Tesla, he did not have that option.

 

[bhzmark]

Discussing ways that tacc can be improved or have added functionality doesnt suggest that the op driver was blameless. I do think that some ACC's are better than others. Tesla is certainly better than the Toyota acc I am used to. Maybe Mercedes and Subaru is better than Tesla with their multiple cameras. And perhaps they would have handled this same situation better. I think that is likely.

 

[dsm363]

Maybe Tesla should disable it below 45mph. I realize stop and go traffic happens on the highway so that would disable it here too.

Simple education about proper use should be all that's needed though.

 

[AnxietyRanger]

I've skimmed through the comments in this thread and I just feel that some main points need to be brought back to the forefront.

• TACC is not autonomous driving. It's cruise control that will attempt to slow for a slower car in front when conditions allow. Conditions don't always allow, as is made abundantly in the manual including specifics to this situation.
• TACC is not designed for use on surface streets. People using it in those areas and getting into accidents or near accidents as a result of their inattentiveness deserve to be in accidents, IMO.
• This accident is 100% the fault of the driver. 100%. Not 99%. 100%. Had the driver paid attention as required in the situation and reacted like any reasonable driver would then there wouldn't have been a problem.

Driver assistance, not driver replacement. Get a grip folks. The OP screwed up, end of story. An unfortunate screw up, but one none the less.

It is quite likely the OP is to blame for this one and TACC acted within legacy ACC limitations like something from a decade ago or from a cheap current car. But frankly, if TACC can't be used on regular streets it is much worse than the German competition. That has to be said. Many German models use ACC for stop-and-go traffic already since many years, meaning the car stops at lights etc. by itself (after the car in front) and even starts driving again by itself if traffic moves again soon enough. I guess Tesla just isn't there yet.

I haven't driven Tesla's ACC so I don't know (I guess I'll see what Model X offers eventually), but I've driven many German ACCs for a decade and they are perfectly usable in most driving situations and very reliable. The latest implementations with multiple radars, camera and combined navigation data (these themselves already around 5 years old) are very impressive in almost any situation. And they've saved my inattentive behind a few times too, by stopping the car - with a stationary car in front.

bhzmark: Agreed. I'm willing to tentatively apportion same "blame" to Tesla's ACC not being very good yet. Not a liability kind of blame, just that it is a limited feature.

 

[wk057]

The point remains, though, and the one I was trying to drive home, that it is the fault of the driver and not Tesla. There is nothing more Tesla was responsible to do in this situation. If we're to say Tesla should provide a delivery orientation for every change of hands and blame them for driver ignorance if they do not... that would be pretty ridiculous as well.

 

[AnxietyRanger]

The point remains, though, and the one I was trying to drive home, that it is the fault of the driver and not Tesla. There is nothing more Tesla was responsible to do in this situation. If we're to say Tesla should provide a delivery orientation for every change of hands and blame them for driver ignorance if they do not... that would be pretty ridiculous as well.

I can believe it was the fault of the driver. Nor do I think Tesla should provide orientation for secondary owners. I think very few of us, the OP included, have continued to discuss legal blame in this thread - except those that have hung up on the point that legal action was even mentioned. Tesla has many defenders here, but nobody seems to be suing Tesla as far as I can tell. Personally I was interested in what happened - and what could have happened differently - from a technical and usability perspective. That gets muddied if one finds a need to defend Tesla in the process, so I try to steer clear of discussing legalities - I doubt anyone of us here need to worry about those for Tesla.

 

[Stoneymonster]

It is quite likely the OP is to blame for this one and TACC acted within legacy ACC limitations like something from a decade ago or from a cheap current car. But frankly, if TACC can't be used on regular streets it is much worse than the German competition. That has to be said. Many German models use ACC for stop-and-go traffic already since many years, meaning the car stops at lights etc. by itself (after the car in front) and even starts driving again by itself if traffic moves again soon enough. I guess Tesla just isn't there yet.

I haven't driven Tesla's ACC so I don't know (I guess I'll see what Model X offers eventually), but I've driven many German ACCs for a decade and they are perfectly usable in most driving situations and very reliable. The latest implementations with multiple radars, camera and combined navigation data (these themselves already around 5 years old) are very impressive in almost any situation. And they've saved my inattentive behind a few times too, by stopping the car - with a stationary car in front.

bhzmark: Agreed. I'm willing to tentatively apportion same "blame" to Tesla's ACC not being very good yet. Not a liability kind of blame, just that it is a limited feature.

What you describe is precisely my experience with TACC on the Tesla.