Another tragic fatality with a semi in Florida. This time a Model 3

[cucubits]

Had it been ANY other car with radar cruise control and some form of lane keep assist, for sure it wouldn't have gotten all over the news. Everyone just love to jump to Tesla's throat whatever happens.

Having said that, the most important thing to take away from this is to make it more clear to everyone how radar works or more precisely how it doesn't with stationary objects.

A good viable solution would be to have a camera pointed towards the driver and have software constantly analyze where he's looking. No eyes on the road, send alarms and slow down. Other manufacturers are already using this, why is this so hard to implement? Is this because of the silly expectation of privacy or some real reason?

 

[StellarRat]

The driver hit near the center of the trailer though (also I believe the trailer was loaded) and again there was ZERO braking. I would be very curious to see your calculations on how this could be an unavoidable accident.
If semis never caused anyone to have to slow down they would never be able to move. In the real world people are forced to yield to semis all the time. Also, you are legally required to avoid accidents even if you have the right of way.

I never said it was unavoidable, I just brought up the possibly because that happens everyday. Personally, I don't use AP any place where there is the possibility of crossing traffic. For me, it's strictly for the freeway.

 

[StellarRat]

Having said that, the most important thing to take away from this is to make it more clear to everyone how radar works or more precisely how it doesn't with stationary objects.

It seems like this problem will have to be solved to ever have a workable FSD. I wonder how Tesla is doing this (or not)?

 

[Daniel in SD]

It seems like this problem will have to be solved to ever have a workable FSD. I wonder how Tesla is doing this (or not)?

Their plan is to use the cameras and deep learning (neural nets). Clearly it's possible to use cameras (the NTSB could clearly see the semi trailer in the cars camera recordings) it's just a matter of software and training the neural nets.

 

[cucubits]

It seems like this problem will have to be solved to ever have a workable FSD. I wonder how Tesla is doing this (or not)?

I'd imagine a new sensor of some sort must be found for stationary obstacle detection. Surely they can't hope to rely on image recognition only.

 

[Daniel in SD]

I'd imagine a new sensor of some sort must be found for stationary obstacle detection. Surely they can't hope to rely on image recognition only.

Like LIDAR or high resolution RADAR? No, they're definitely planning to do it with image recognition only.
The argument being that if humans can do it then so can machines.

 

[mongo]

No, we know that no torque was detected on the steering wheel for a few seconds before the accident. No one has any idea where the driver's hands were.

However, we do know the hands did not try to avoid the accident by turning the wheel. Even if they were on the wheel, they didn't do their job.

 

[Eno Deb]

Had it been ANY other car with radar cruise control and some form of lane keep assist, for sure it wouldn't have gotten all over the news. Everyone just love to jump to Tesla's throat whatever happens.

True, but then these other carmakers don't hype up and sell "full self-driving" ...

A good viable solution would be to have a camera pointed towards the driver and have software constantly analyze where he's looking. No eyes on the road, send alarms and slow down. Other manufacturers are already using this, why is this so hard to implement? Is this because of the silly expectation of privacy or some real reason?

For one, it seems Elon is opposed to it, and second, the interior camera (which is reportedly meant to monitor the passengers of Robotaxis) may not have the right position and zoom factor to reliably detect the gaze of the driver. There is also no interior camera in Model S/X.

 

[Eno Deb]

Their plan is to use the cameras and deep learning (neural nets). Clearly it's possible to use cameras (the NTSB could clearly see the semi trailer in the cars camera recordings) it's just a matter of software and training the neural nets.

You say that as if it was the easy part. But in reality it's really really hard, and some researchers have doubts if it will ever be reliable enough without a major breakthrough.

 

[Daniel in SD]

You say that as if it was the easy part. But in reality it's really really hard, and some researchers have doubts if it will ever be reliable enough without a major breakthrough.

Haha. I'm as big a FSD skeptic as anyone here. It's ridiculously difficult and I don't see how anyone can have timelines on technological breakthroughs. I was just saying that's what Tesla's plan is.

 

[Cloxxki]

I read about htem. cannot reproduce myself. Also it's very possible those have nothing to do with radar and are just a code bug.

Nice to blame it on radar, but that's the tool Tesla chose and is sticking with. I can't say I'm impressed. Should be pretty elementary to detect relevant object before starting a software campaign to make a car independent of human supervision.

 

[SMAlset]

They said he didn't turn on AP until shortly before the accident occurred, so he drove his morning trip manually for the most part. For whatever reason he felt at this juncture he could use AP and he didn't pay full attention to the road (this wasn't a new road to him but one he had traveled before). My guess is he turned it on and got distracted (looking at email, texting, resting his eyes, spilled his coffee, who knows and don't believe we will ever know) and didn't look up in time. The fact the report said the truck slowed down after it pulled out to me sounds like that driver didn't look to see oncoming traffic in that section of the highway (hard to believe he just didn't care if he got hit). Like who doesn't step on the accellerator when you see yourself about to be hit to try at least to clear out of the way. The slowed trailer may have also been part of the reason on AP why he might not have had a warning from the car. In any event he shouldn't have been using AP on a divided highway for exactly this reason. One Florida accident with a car on AP should have been a life lesson for drivers using it on divided highways, now we have two.

Their plan is to use the cameras and deep learning (neural nets). Clearly it's possible to use cameras (the NTSB could clearly see the semi trailer in the cars camera recordings) it's just a matter of software and training the neural nets.

Fact is people make mistakes. And no matter how much neural net training one does there's still the factor that human's aren't always predictable so neural net can't be either, might have a higher percentage of predicting an action but there will always be room for that exception. Telsa and other car manufacturers know accidents and some fatal will still occur, but hopefully the technology will reduce the numbers. Even from accounts people here have made there have been "saves". Just can't save them all. This is still something I see the public and press not understanding fully. I don't see news stories for the most part being presented in a beneficial way to help people understand the present limitations, and instead see it fightening people off or on the otherhand sometimes promoting unintended usage.

@Oldschool496, wow that must have hit you like a ton of bricks. Sorry you and your family lost someone you knew in this fashion. Thank you for sharing that with us. Helps I think make it more real.

 

[SMAlset]

Nice to blame it on radar, but that's the tool Tesla chose and is sticking with. I can't say I'm impressed. Should be pretty elementary to detect relevant object before starting a software campaign to make a car independent of human supervision.

You can add Nissan to that approach now. On the radar: Nissan stays cool on lidar tech, siding with Tesla - Reuters

The fact that Tesla says don't use AP on divided highways is kind of paramount right now. The neural net technology isn't there yet to rely on, although you see drivers' videos out on the internet misusing AP and essentially promoting its misuse.

 

[MXWing]

“Protect yourself at all times”.

I don’t give a F what Musk says, my friends say, the release notes say.

“Protect yourself at all times”.

I’ve looked at email, texts, sent texts while on AP I’ll admit.

Though it takes me a long time because 9/10 cycles is spent on the car/road not 1/10 which appeared to have been the case here.

 

[Barklikeadog]

That's not true. An unloaded semi can move pretty quickly. If you're in the right lane you can be toast. However, I do agree with rest of your post. I also think the semi failed to yield which is something no one seems to be mentioning here.

Like the poster above you stated, AP has a problem with non/slow-moving objects. I would think the radar would be useful, but as I've stated in other posts a Doppler radar is made for detecting moving objects and I have no idea how capable it is with stationary objects or if Tesla uses it in that mode if available.

Apparently AEB and frontal collision warning systems have a problem too. And this is exactly what they are designed for.

 

[mongo]

Apparently AEB and frontal collision warning systems have a problem too. And this is exactly what they are designed for.

They are primarily designed for objects in the driving path, not for limbo.

 

[Barklikeadog]

They are primarily designed for objects in the driving path, not for limbo.

no worries, full self driving is coming this year.
sleep in your car or robotaxi next year

 

[cucubits]

I'd like to see them repeat their FSD demo (like where they had the car drive itself around town, through intersections, etc) but with a 'planned' unexpected obstacle. Like have a huge cardboard box pull in front of it...

Honestly they did more bad than good with that video. There must be some people with teslas out there who will try to drive like that...

 

[Silicon Desert]

Barring a medical condition, Just judging from the distance the car traveled after the collision he was at highway speed.

Unless I got this story confused with another one, I recall he was going 68 MPH in a 55 zone. It doesn't help to be speeding.

 

[afadeev]

Lex Fridman's work at M.I.T showed that Tesla drivers are more attentive while on autopilot

Intuitively, and based on personal experience, this assertion makes NO SENSE.
If this is the study in question, you misunderstood or misrepresented its conclusions:
Human Side of Tesla Autopilot | MIT

If it's another study, please provide the link to validate the claim.

They will be able to with the new software which hasn't shipped yet--only been demonstrated.

Do you have any evidence to support this claim?
We've all read similar hopeful pronouncements after the last guy drove his Model S under a semi-trailer and decapitated himself ~3 years ago.
Yet, here we are again.

Tesla has always been clear that the driver needs to pay attention to the road at all times and keep hands on the controls. Driver in this case did not.

That is very true, but the very name of the product, AutoPilot, runs contrary to this declaimer.
If the product was named "Lane keep assist with dynamic cruise control", everyone would have much lower expectations and trust in its capabilities.

Yes, Tesla's AP works better than an average industry standard LKAS with DCC, but perfect it is clearly not.
AutoPilot - that is just a misnomer.

I drive a Tesla obviously, these cars even with Autopilot do not stop for objects in the roadway stopped or I guess in this case barely moving. IF they did they would never move. They would stop for everything the camera(s) see. You (we) must understand this. WE are still driving the car. We are still responsible for our safety and of others.

Its a wake up call to us all. IF you think this system is 100% going to do what you do with full attention in its current configuration and no FSD, your in dream state here. Your still driving this car. YOU the person behind the wheel, YOU.
I hope I never forget that. In the early years of ownership I was misled by my own ideas, I read and read and read again, we are driving these cars, these systems are just like cruise control in current configuration.

Sadly, true.

a