Tesla 3 crashes into overturned truck on highway

[diplomat33]

This happened earlier today in Taiwan. A Model 3 on the highway crashed head on into an overturned truck.

A surveillance camera caught the whole thing:

Apparently, the reports say that some AP driver assist features were on.Sounds like maybe just TACC was on and not Autosteer.

This is why the driver needs to always pay attention. Driver assist is not FSD.

 

[Watts_Up]

Most important the driver didn't get injured.

Reports said that the driver was sleeping?

The official website I read shows that Tesla in Taiwan only has "automatic assisted driving".
It may be that the driver mistakenly thinks that it is "automatic driving" without paying attention
to the road conditions. It may also be that the driver is asleep​

Also reports said that the car didn't slow down, but at 0:10 you can see smoke from the tires?

- So does the Tesla detected the truck or does the driver noticed the person on the side at 0:54 to warn to slow down?

Even though at 0:18 and 1:02 you can see a Mercedes not slowing down until 0:26 and 1:11
where you can see the smoke from the tires and almost hitting the Tesla.

Anyway, this guy on the side, may be the driver of the truck, is an hero!
Without him, a pile up would had occurred.

 

[diplomat33]

Most important the driver didn't get injured.

Yes, that is wonderful news.

 

[Tam]

...not Autosteer...

It's dead centering the lane to me.

If Autosteer was not on, I would expect some deviation toward left or right from the center of the lane escpecially for an inattentive driving that doesn't even notice a big gigantic truck in front.

 

[diplomat33]

It's dead centering the lane to me.

If Autosteer was not on, I would expect some deviation toward left or right from the center of the lane.

True. I am just reporting what I read that "some driver assist features were on". You are right that autosteer might have been on.

 

[Tam]

...at 0:10 you can see smoke from the tires?

- So does the car detected the truck or does the driver noticed the person on the side at 0:54 to warn to slow down?

The radar detects all objects: benign harmless ones as well as the dangerous ones.

The issue has always been how the radar can discriminate which is which so it can brake appropriately and that's why so many drivers are complaining phantom brakes.

It's possible that the automation did apply the brakes but not in time to stop a collision.

In my case, The TACC (Traffic Aware Cruise Control) did the initial braking and deceleration as evidenced by lit up both teal icons which mean the automation was not disabled, lit up brake lights, and a speed reduction from 61 to 53 MPH but not hard enough to avoid a collision. Thus, I had to manually take over the braking which disabled the automatic system at that point.

 

[MXLRplus]

That's not tire smoke from braking. That's some kind of white powder on the roadway.

 

[Watts_Up]

特斯拉撞穿大貨車！「保命關鍵」曝光…駕駛靠這招毫髮無傷 | 社會 | 三立新聞網 SETN.COM

Tesla crashed into a big truck! "Life-saving key" exposure ...
Driving on this trick without any harm | Society | Sanli News Net SETN.COM

 

[Mardak]

I wonder if Tesla's pseudolidar depth prediction would have detected the truck. Although a large white surface probably makes it more difficult, but even ignoring that, it looks like the examples with main camera don't predict depth too far away:

Notice in the left, the overhead roadway and supporting columns are only very faintly predicted to be closer. And the middle example has a crossing bus that is again only very faintly closer. The right example is just showing that closer objects are better highlighted even with a mostly white color. Although potentially this deficiency is just an artifact of the visualization process and maybe the predictions are confident enough to change driving behavior.

Also Karpathy showed off depth predictions with the wide and pillar cameras, so having the neural network also calculate depth with the narrow (250m > 4x distance of wide's 60m) could help for highway speeds.

 

[Tam]

That's not tire smoke from braking. That's some kind of white powder on the roadway.

Very observant.

Later on the clip, it also shows the angle for us to observe the red brake lights from the rear of cars. There was none for the Tesla even when the white plume started to fly off.

We could see brake lights for other non-Tesla cars.

 

[Watts_Up]

On this video, after showing the Tesla hitting the truck,
at 2:20 there is a black Toyota hitting a bus after been in the wrong lane,
but there is no mention of it in the news web page?

畫面曝光／特斯拉變身穿雲箭撞進貨車！一陣煙…半台車消失
Screen exposure / Tesla turned into a cloud arrow and hit the truck!
A burst of smoke ... Half a car disappeared​

 

[Watts_Up]

Very observant.

Later on the clip, it also shows the angle for us to observe the red brake lights from the rear of cars.
There was none for the Tesla even when the white plume started to fly off.

We could see brake lights for other non-Tesla cars.

On the video the red bake lights from the trucks are noticeable but are very difficult to spot on the cars.
The Mercedes might have the red brake light on the right side turned only at 1:09.

 

[Watts_Up]

It seems that the driver air bag didn't deploy?

 

[P3dStealth]

This guy was totally not paying attention.

What's concerning to me though is why are Teslas still not even slowing down for giant solid objects in the way? I feel like they should have figured out the Tesla driving into big trucks problem by now.

 

[Yang_H]

I remember that it has been discussed many times here, that all current autopilot or kind likes ignore stationary objects once moving above the speed of ~50mph, due to some kind of tech limitations. So unfortunate.

 

[Morristhecat]

Love how everyone else just keeps driving by, as if nothing to see here... Nobody stops to help the truck driver or the tesla driver/passengers. What's wrong with everyone? Just merge over and keep going....

 

[diplomat33]

I remember that it has been discussed many times here, that all current autopilot or kind likes ignore stationary objects once moving above the speed of ~50mph, due to some kind of tech limitations. So unfortunate.

Yes because driver assist that rely on front radar have to ignore stationary objects because otherwise there would be too many false positives. So you have to deliberately tune out all the stationary objects, or the car would be braking for everything. But this means, that the car will also ignore real obstacles like an overturned truck. But this also means that any car that relies on this system, can never be FSD because the driver will always have to pay attention for these false positives.

 

[aronth5]

Yes because driver assist that rely on front radar have to ignore stationary objects because otherwise there would be too many false positives. So you have to deliberately tune out all the stationary objects, or the car would be braking for everything. But this means, that the car will also ignore real obstacles like an overturned truck. But this also means that any car that relies on this system, can never be FSD because the driver will always have to pay attention for these false positives.

So is your assumption that because that is how FSD works now it will always work this way? That seems a bit of a stretch.

 

[diplomat33]

So is your assumption that because that is how FSD works now it will always work this way? That seems a bit of a stretch.

No, I am not saying that at all.

I am saying it is not FSD yet because it works this way. Being able to respond to stopped obstacles on the road is a prerequisite of FSD. Obviously, if it changes, it can become FSD.

There are basically 2 ways to solve this problem in order to do FSD:

1) You train the camera vision to accurately detect stopped obstacles.
2) You use another sensor like lidar that is able to accurately detect stopped obstacles.

Tesla is trying #1 but clearly, they are not there yet. Until, they can do #1, AP will not be FSD.

 

[Tam]

So is your assumption that because that is how FSD works now it will always work this way? That seems a bit of a stretch.

That is how the current radar system works and for a foreseeable future, it will always work that way until someone is smart enough to teach radar to differentiate between benign objects and dangerous objects.

As @diplomat33 mentioned above, to compensate for that, Tesla could use Tesla Vision but it is not advanced to this kind of function just yet or

Lidar proponents say while waiting for radar and camera vision to perfect, they can use lidar in addition.