AP Crash Los Angeles, protruding bridge column, Model 3

[DanCar]

12 seconds into the video, Autopilot apparently didn't see the protrusion from the bridge. This is a heavily traveled and one of the major freeways in Los Angeles. Given there were no shoulders, the margin of error were very low. The crash pushed my vehicle into the other lane. Luckily, I didn't hit another car. I've been using autopilot for almost 3 years with no problems. After this incident, it's likely I won't be using autopilot again on roads with no shoulders. Tesla informed me that it'll take 3-5 weeks for them to analyze the autopilot data and get back to me. Meanwhile, I'll need to get my car fixed with my own insurance. Comments are turned off but is available in the Tesla Model 3 forum under Model 3 Autopilot accident. Model 3 Autopilot accident

Other forum:
https://www.reddit.com/r/SelfDrivingCars/comments/lvi5dy/tesla_model_3_autopilot_crash_feb_27_2021/

 

[MXLRplus]

I've been telling people that AP/FSD does not react fast enough. Especially when avoiding changes in the K-Rail or center divider.

DO NOT use AP/FSD if there is no emergency lane and there is K-Rail. You might miss 99% of the time, but it will hit eventually. Just observe WHEN the car starts to turn. It is over one car length too late. This system is not going to be full self driving unless they fix it's reaction time.

Neither Jaguar or Cadillac has this problem, so it's not an impossible situation.

 

[MXLRplus]

It will be funny to see the first post from somebody claiming that's a desirable feature. It apparently saved the life of an endangered blunt-nosed lizard who was on the dotted line.

 

[Two-rocks]

The blunt nosed lizard was in the pot hole - that the AP seems to seek out.

 

[Uncle Paul]

For the most part, I tend to use autopilot when cruising in the middle freeway lanes. Driving in the far right causes issues with entrance and exit lanes with poor markings and merging traffic. Hugging the fast lane often gives zero margin for error in case of debris or poor design.

This is not only for Autopilot, but also when I am manually driving. I just do not like riding on the outside lanes when there is little margin for error.

 

[powertoold]

I'm curious why his camera quality is so poor. I can barely see the protrusion myself.

 

[thewishmaster]

I'm curious why his camera quality is so poor. I can barely see the protrusion myself.

A combination of low contrast due to the position of the sun and YouTube compression, probably

 

[nomis_nehc]

Other forum:
https://www.reddit.com/r/SelfDrivingCars/comments/lvi5dy/tesla_model_3_autopilot_crash_feb_27_2021/

I think I know where this is. You were going east, yeah?

Hate to say this, but you're supposed to be watching the road, I always manually steer away just in case when I pass that part.

 

[powertoold]

Judging from the markings on that protrusion, it looks like it's happened to at least 2 other cars:

 

[run-the-joules]

Yup, I've seen things like this and I've experimented with waiting as long as I'm willing to see if the car will react, and more often than not I lose the game of chicken and take over.

 

[Dan D.]

Judging from the markings on that protrusion, it looks like it's happened to at least 2 other cars:

Agreed. All other things considered is it really so hard to just do something helpful?

 

[richrootes]

That shouldn’t even be a travel lane

 

[DanCar]

This is proof tesla needs lidar! . . . . . haha just trolling.

 

[acentre]

crazy California road design, but the hit was operator error!

 

[Tam]

...the hit was operator error!

I think by now, most would realize that the driver is responsible and the driver didn't ask Tesla to pay for the damage. He fixed the damage using his own insurance.

So, denial of driver's responsibility is not the issue here.

I think the issue here is to learn how the system allows the accident to happen and how to fix this particular Autopilot deficiency.

I think the current Autopilot is not coded for this scenario: The radar, the cameras, the sonars do record that there's a narrowing of the lane but no one in Tesla has coded that the system should also react appropriately and avoid hitting the left wall.

Nothing is wrong with the sensors but plenty is missing in the programming codes.

It's unknown how long will this problem be solved or has it been solved with the current FSD beta (limited released version to a few).

If it's not solved with the current FSD beta then I think adding LIDAR should be considered.

 

[MP3Mike]

and the driver didn't ask Tesla to pay for the damage.

Actually it seems like they have.

 

[DanCar]

Actually it seems like they have.

What makes you say that?

 

[Tam]

Actually it seems like they have.

The quote from Tesla forum:

"I'm not assuming nor expecting Tesla to pay for the damages but I do hope they own up to it and do the right thing. They did tell me that if it's determined to be an autopilot issue, they will pay for the damages..."

 

[MP3Mike]

What makes you say that?

In addition to the quote Tam provided there is this one: "Tesla informed me that it'll take 3-5 weeks for them to analyze the autopilot data and get back to me. Meanwhile, I'll need to get my car fixed with my own insurance. "

Making it seem like he asked Tesla to cover the cost and they will let him know in 3-5 weeks if they will or not.

 

[dgatwood]

It will be funny to see the first post from somebody claiming that's a desirable feature. It apparently saved the life of an endangered blunt-nosed lizard who was on the dotted line.

No, it's a Gekko. And yeah, it will pay his salary.