"Obstacle detected" - blocking acceleration - very dangerous!

[S4WRXTTCS]

I have to point out, as I have before, that there are always gonna be a bunch of roads which aren't really 'designed' at all and aren't very compatible with any sort of driver aids. Dirt and gravel roads are the most obvious, but lots of ancient rural roads are like this, as well as ancient streets in old cities like Boston. It's just a wild and wooly environment. This is why I've been saying that full level 4 self-driving is impossible.

I don't really think going straight to level 4 self-driving is possible either. What I do believe is possible for a car to self drive over specific approved white listed roads (on-grid driving). Where the number of roads that are white listed grows every month. The white list would grow every month from both technical improvements to the cars, but also infrastructure enhancements.

The same thing has to happen with active safety systems. There are some active safety systems that are simply not compatible with the infrastructure (like what Tesla plans on doing in a future version of 8.X). So they'll have to have there own white list at least until the technology evolves to be 99.9% effective if not higher.

I don't have any evidence that other cars have had issues with this specific road section, but at the same time I would find it hard to believe that there aren't. I need to do some research to see if there is anyone logging road sections where active safety equipment have had false triggers. Either AEB or throttle control (like what Subaru does in addition to AEB). It is extremely dangerous, and turning it off (when given the option) defeats the entire purpose of it.

 

[Colby Boles]

Just to update everyone I haven't heard back anything conclusive from Tesla yet but I have been told that the the Director of Autopilot Systems will contact me directly, so I it does at least sound like the issue is being taken pretty seriously.

 

[MP3Mike]

Thanks for the update!

 

[S4WRXTTCS]

Just to update everyone I haven't heard back anything conclusive from Tesla yet but I have been told that the the Director of Autopilot Systems will contact me directly, so I it does at least sound like the issue is being taken pretty seriously.

Too bad we can't have a questions and answers online session with that guy.

Here would be my questions.

1 - Why isn't "Obstacle detected" based throttle cutoff mentioned in the manual?

2 - Is truck lust an expected phenomenon with AP? Why does it seem to occur more right after a major update (from 7.0 to 7.1, and 7.1 to 8.0)?

3 - Do AP tiles get purged during an AP related update?

4 - How are AP tiles managed? Are they cached on the car or downloaded as you go?

 

[bhzmark]

Looking at the video again and considering my similar experience I'm pretty sure the radar or camera is sensing the side barrier on the right as the obstacle and if the vehicle were oriented pointing more clearly away from the side barrier that it shouldn't detect the sidewall as an obstacle.

Try that on ramp again pulling the vehicle far to the right side of the lane against the wall but then pointed more away from the wall into the new Lane of traffic that you intend to go so that there's less chance the radar or camera will see the side wall as being in front of it and an obstacle

 

[Colby Boles]

Looking at the video again and considering my similar experience I'm pretty sure the radar or camera is sensing the side barrier on the right as the obstacle and if the vehicle were oriented pointing more clearly away from the side barrier that it shouldn't detect the sidewall as an obstacle.

Try that on ramp again pulling the vehicle far to the right side of the lane against the wall but then pointed more away from the wall into the new Lane of traffic that you intend to go so that there's less chance the radar or camera will see the side wall as being in front of it and an obstacle

I have yet to have the opportunity to pull on to this freeway when no cars are present. It has been on my list of things to try though as it would resolve the stationary object theory you bring up. I don't think it is the barrier only because the car is pointed straight out into the other lane and no turning is needed to avoid the barrier. Of course the car could be thinking otherwise. Once you have accelerated into the lane, you actually need to turn right a bit to avoid crossing more lanes. If I can reproduce the issue with no cars present I will know you are on the right track.

Unfortunately it takes a lot of driving to loop back to this same on-ramp, otherwise I would try this over and over to figure out exactly what is going on.

 

[bhzmark]

I have yet to have the opportunity to pull on to this freeway when no cars are present. It has been on my list of things to try though as it would resolve the stationary object theory you bring up. I don't think it is the barrier only because the car is pointed straight out into the other lane and no turning is needed to avoid the barrier. Of course the car could be thinking otherwise. Once you have accelerated into the lane, you actually need to turn right a bit to avoid crossing more lanes. If I can reproduce the issue with no cars present I will know you are on the right track.

Unfortunately it takes a lot of driving to loop back to this same on-ramp, otherwise I would try this over and over to figure out exactly what is going on.

hmm. I just don't think the obstacle detection can be that stupid to think a car moving away in front and picked up on doppler radar which makes clear that it is moving fast away should be classified as an obstacle on a potential collision course. the doppler radar should not make that false positive. Perhaps the camera would by taking a single snapshot, but it ought to take and compare multiple images to assign a collision/ no collision prediction to the potential obstacle.

I think it more likely that the radar or esp the camera might sense the concrete wall as somewhat in front and predict that as a collision course obstacle.

 

[Colby Boles]

I should have an opportunity to test this tonight when I use the ramp again. For this try I am going to keep the car oriented the same way as the video but only pull out when there are no cars around (assuming that opportunity arises). If the car still triggers, then in a couple days I will try to point the vehicle more perpendicular to the traffic flow and see if that fixes the problem.

 

[dbsynergy]

I live on Treasure Island and regularly use that exact ramp on my daily commute. I have never had the car act strange there or decelerate me. Maybe issue might be unique to your car? I don't know.

Hopefully Tesla can address the issue in your car. Seems like something is very wrong.

 

[Chickenlittle]

I just had something very scary and dangerous happen less than two hours ago that hopefully never occurs again. I live in San Francisco and I was visiting a storage facility on Treasure Island. I was returning to San Francisco by merging via a stop sign directly on to I-80 West. For people who know this freeway entrance, it is very abrupt and you need to go from 0-60mph very quickly to merge into oncoming traffic that is coming out of the Yerba Buena Island tunnel. This of course is no problem for the P90DL, or a number of other cars I have driven on this route before.

So I see a more than adequate opening between cars coming up, wait for the first car to pass, and then step on the accelerator from my dead stop. For a fraction of a second the car begins to accelerate, and then nothing. I step on the gas several more times very quickly as the car has now rolled out into the lane with a car closing on me at 60+ mph, and still nothing happens. I hear a warning sound and look down at the console and there is a warning symbol and it is telling me that there is an "obstacle detected", and I can see that the right front corner sensor is at the whitish yellow level - which is not even close to anything by San Francisco driving and road width standards.

The second car swerves to the next lane over and whizzes by me. After a couple of seconds the accelerator seems to work again and I am able to drive away.

This car disabling the accelerator because it thought it might be close to something has to be one of the worst and most unsafe UI choices I have ever seen. I could have easily been rear ended as well as injured the driver behind me. I would much rather have the car let me drive into it's imaginary wall than ignore my inputs at such a critical moment.

I called support and Tesla has the logs now, so hopefully they can get to the bottom of what happened. Otherwise, everything seems normal about the car, but I may have some trust issues with it for a while...

Reminds me of a suburban we had around 2005. Twice car shut down with message shutting down to avoid damage to engine once I similar situation entering highway and second time crossing a high speed road with us blocking two lanes. Each time a faulty sensor triggered it. It should be left ultimately to the driver whether to continue despite the perceived risk to the car (especially if risk is to the engine and not life). I liked tesla response to imminent fire risk to have driver pull over and get out rather than car just shutting off on own

 

[Colby Boles]

I live on Treasure Island and regularly use that exact ramp on my daily commute. I have never had the car act strange there or decelerate me. Maybe issue might be unique to your car? I don't know.

Hopefully Tesla can address the issue in your car. Seems like something is very wrong.

Have you left the stop sign at full throttle? I also don't have issues if traffic is slower and I pull in with moderate throttle. This "feature" only seems to trigger under heavy acceleration for me.

 

[dbsynergy]

Have you left the stop sign at full throttle? I also don't have issues if traffic is slower and I pull in with moderate throttle. This "feature" only seems to trigger under heavy acceleration for me.

Now that you mention it. In my last car I would sometimes have traction issues and traction control would kick on when driving over the metal non-concrete parts of the bridge where the bridge sections connect. I think I have had that happen in a couple times in my model x.

Edit: I just rewatched your video, it slows you down right when your front wheels drive over the metal grate just passed the stop sign. My bet is it is your traction control stopping the launch.

 

[S4WRXTTCS]

Traction control doesn't show that message though.

 

[dhanson865]

An update after taking to support at the service center several times who is in turn communicating with other tech support within Tesla who have looked over the logs multiple times:

• There is not enough detail in my logs without any accompanying video (I haven't installed my BlackVue camera yet) to know exactly why the sensors detected what they did, but they seem to think the systems were operating "normally".
• The logs do not say which sensors are activated, just that an impending collision was detected at a system level.
• It is likely that the sensors detected the first car that drove past me and since neither the ultrasonic or radar sensors are Doppler based, they only know that an object is close at some point, not it's trajectory.
• Since I stepped on the accelerator shortly after this car went by, the car likely decided that it would cause a collision if it accelerated quickly based on its limited sensor information that may have latency and does not account for movement of said object.
• It was confirmed that once this "driver assistance" feature triggers on an acceleration based impending collision, it disables acceleration inputs for some seconds afterwards.
• This feature is not related to the "Collision Avoidance Warning" and cannot be turned off.
• Perhaps this "feature" will be refined, improved, or be optionally disabled in future releases, but no-one knows at this point.

Overall, my support person has tried to be as helpful as possible, but the overall message is that this is normal, intended behavior and there is nothing I can do about it at this point. Needless to say I'm pretty concerned. I have only put a few hundred miles on this vehicle and I have already had something happen that hasn't happened to me in 30 years of driving. Thankfully I have never stalled a vehicle at such a critical moment until now. I have also only used this on-ramp twice with the MX and have a 50% failure rate. I did what I considered to be a not very aggressive maneuver with the vehicle (I did not try to accelerate in a way that I could have possibly hit the first vehicle, even with a car twice as quick - I waited for it to pass at ~60mph), and the vehicle left me stranded in a dangerous situation. I can imagine similar scenarios in San Francisco right on my block where one could pull out into a fast-moving one-way street right after a car passes and others are coming, either from a parking spot, driveway, or cross-street where the trigger conditions would be similar and you would be at risk if you can't get the vehicle moving quickly.

I do find it interesting that no-one else has reported this problem though. Maybe it is truly a very rare occurrence, but I'm a bit skeptical considering how few driving miles I have put in so far. I'd like to go back and reproduce the problem some more (minus the second vehicle behind me needing to make an evasive maneuver), but it is a huge PITA at this location because one must traverse the bridge span, downtown areas, and Treasure Island on and off-ramps each time you wish to repeat the scenario.

I'd love to hear from anyone else who is able to reproduce this sort of thing, but at the same time be very careful doing so as the car is disabled for much longer than makes any rational sense.

I'd report to NHTSA based on this content alone. Obviously I won't report it as I don't own your car nor did I have your experience but if I were you I would have already done so.

I believe TSLA deserved a chance to fix first but the second they say you can't fix it and it's normal behavior (or alternatively if they fixed it and said X caused the issue) either way you have enough data to make a valid report.

 

[dhanson865]

Every car that has traction control does that. Where if the wheels spins it brakes that wheel, and reduces the acceleration.

My Jeep does the same thing, but I manage it by knowing when to feather the throttle. It's mostly when it's just rained after not raining for awhile.

I believe the Tesla does it as well, but it's extremely difficult to lose traction in the Tesla.

Basically the more advanced the drive by wire/ABS/autopilot/whatever gets the more important it is that you buy better quality tires and keep them inflated properly.

Traction expected better match traction experienced or odd things happen.

 

[Colby Boles]

I'd report to NHTSA based on this content alone. Obviously I won't report it as I don't own your car nor did I have your experience but if I were you I would have already done so.

I believe TSLA deserved a chance to fix first but the second they say you can't fix it and it's normal behavior (or alternatively if they fixed it and said X caused the issue) either way you have enough data to make a valid report.

In some of my other posts in this thread I mention that I did report this to the NHTSA. I haven't heard anything back from them though - I should look into that.

 

[krouebi]

Hi,

now there's a really weird and dangerous freeway entrance!

If there are no vehicles in front, can you see the traffic coming out of the tunnel when keeping to the left of the entering road?

This would give you the possibility of a longer acceleration distance, so you would not have to use maximum acceleration.

Does the problem occur if you do not accelerate fully? Then don't - 80% is still VERY fast.

Does the problem occur if you do not come to a complete stop? Then roll through.

IMHO the problem lies more with the way the entrance is laid out - a fully loaded VW Bus (the original one) is allowed on this entrance, but will NEVER be able to merge safely.

 

[apacheguy]

Reading your description of the issue and the warning message suggests to me that your right front corner ultrasonic sensor malfunctioned. Have the Burlingame Service Center check it out.

Sure, but if AP is to succeed by the masses Tesla will need redundancy so that a single faulty sensor can't cause an accident as it could have here.

 

[Colby Boles]

* You can see the cars coming from the tunnel. The issue is that this is a very busy interstate highway and much of the time all lanes are full of cars going 60mph - you might be waiting for an opening for a long time if your car is incapable of accelerating quickly.

* I don't know specifically about 80% throttle, but at something less than full acceleration, I have not noticed the problem. Yes, if you know about this limitation ahead of time at this location you can avoid the issue. The bigger problem is that you don't know when the car is going to do this, and when it does, you end up with the equivalent of a stalled vehicle right at the time you needed to accelerate most. 80% pedal gets you ~80% acceleration - 100% pedal gets you either 100% or 0% + some rollout just to make it extra dangerous.

* You need to stop here to assess the situation.

* There is no reason the car should do this. The radar sensor supports Doppler, so it should know that the passing cars are moving quickly, and even if they are a few feet away the car could accelerate quickly with no risk of collision.

Hi,

now there's a really weird and dangerous freeway entrance!

If there are no vehicles in front, can you see the traffic coming out of the tunnel when keeping to the left of the entering road?

This would give you the possibility of a longer acceleration distance, so you would not have to use maximum acceleration.

Does the problem occur if you do not accelerate fully? Then don't - 80% is still VERY fast.

Does the problem occur if you do not come to a complete stop? Then roll through.

IMHO the problem lies more with the way the entrance is laid out - a fully loaded VW Bus (the original one) is allowed on this entrance, but will NEVER be able to merge safely.

 

[jobrien70]

I should have an opportunity to test this tonight when I use the ramp again. For this try I am going to keep the car oriented the same way as the video but only pull out when there are no cars around (assuming that opportunity arises). If the car still triggers, then in a couple days I will try to point the vehicle more perpendicular to the traffic flow and see if that fixes the problem.

Have you been able to try this merge zone without any traffic? It would be great to help isolate what may be triggering the throttle to cut off.