Tesla On Autopilot Slams Into Stalled Car

[nvx1977]

Given his credentials, the author knows this. Why he chooses to write such a poor article is a mystery to me.

Because if you dig into his credentials, particularly where he's investing his money, he's likely backing LIDAR-using companies.

I'm certain that this guy is fully qualified to write a very well-informed article. There are elements of that in the article in question. He just lets his bias show on the LIDAR bit. Look up his contribution on Forbes and you'll see he turns way more negative on Tesla after the Autonomy Investor Day presentation.

I personally don't care if he's for or against Tesla. I just want to know where authors are coming from so I can normalize the article for any bias. I do the same for the electrek and teslarati fanboy articles as well.

 

[PoitNarf]

The question is about recognizing stationary objects.

My car does it just fine.

My car recognizes stopped cars in front of it just fine at intersections.

My car recognizes the back wall in my garage just fine when it parks itself.

My car recognizes me in front of it during summon.

I'm not sure what people are saying about it not recognizing stationary objects.

Does anyone want to amend their statement about Tesla's not recognizing stationary objects?

Yeah the radar detects stationary objects just fine. However the software is designed to ignore most of the stationary objects otherwise you’d experience constant phantom braking events. This is why these sort of accidents happen, the car is purposefully ignoring stationary objects reported by the forward facing radar. From what I’ve read this is how all radar assisted cruise control systems work. This isn’t a Tesla specific problem, it’s a shortcoming of radar assisted cruise control and any vehicle from any manufacturer that utilizes this would probably collide with the stationary object in its lane.

Eventually Tesla should be able to properly utilize the data from the cameras to help avoid similar collisions in the future.

Why emergency braking systems sometimes hit parked cars and lane dividers

 

[Garlan Garner]

Because if you dig into his credentials, particularly where he's investing his money, he's likely backing LIDAR-using companies.

I'm certain that this guy is fully qualified to write a very well-informed article. There are elements of that in the article in question. He just lets his bias show on the LIDAR bit. Look up his contribution on Forbes and you'll see he turns way more negative on Tesla after the Autonomy Investor Day presentation.

I personally don't care if he's for or against Tesla. I just want to know where authors are coming from so I can normalize the article for any bias. I do the same for the electrek and teslarati fanboy articles as well.

Everyone,

Tesla isn't going to use LIDAR period.

No matter how many articles are written about it.

 

[Garlan Garner]

Yeah the radar detects stationary objects just fine. However the software is designed to ignore most of the stationary objects otherwise you’d experience constant phantom braking events. This is why these sort of accidents happen, the car is purposefully ignoring stationary objects reported by the forward facing radar. From what I’ve read this is how all radar assisted cruise control systems work. This isn’t a Tesla specific problem, it’s a shortcoming of radar assisted cruise control and any vehicle from any manufacturer that utilizes this would probably collide with the stationary object in its lane.

Eventually Tesla should be able to properly utilize the data from the cameras to help avoid similar collisions in the future.

Why emergency braking systems sometimes hit parked cars and lane dividers

Wait wait....hold on.

The car DOES NOT ignore stationary objects in front of it. Lets get that clear.

The statement I was responding to was this: The car does not recognize stationary objects.

Conclusion: It does recognize stationary objects.

--------------------------

Now...my radar/camera based NoA has been flawless. It recognizes both moving and stationary cars in front of me daily. AND it does it better than it did a month ago....which makes me wonder why articles and videos from a year ago are valid in this discussion.

 

[Wennfred]

Great topic guys, and as far as Stats, ICE have been on the market for what the 1800’s? And the EV evolution since probably 2000 +/~ 10 years. So for sure there are more ICE cars slamming into other cars since day one.

Now If only we can get Tesla to fix the Auto Laundromat mode, maybe I might get AP.

Fred

 

[Garlan Garner]

Great topic guys, and as far as Stats, ICE have been on the market for what the 1800’s? And the EV evolution since probably 2000 +/~ 10 years. So for sure there are more ICE cars slamming into other cars since day one.

Now If only we can get Tesla to fix the Auto Laundromat mode, maybe I might get AP.

Fred

What does that mean Fred? What is Auto Laundromat?

 

[C141medic]

It may stop for rabbits and geese

Watch how Tesla Autopilot sees a goose crossing the road - Electrek

 

[Garlan Garner]

How about someone post..

How many non Tesla ( ICE ) cruise control accidents happen every day.
How many non Telsla ( ICE ) drivers fall asleep driving and have accidents every day
How many non Tesla ( ICE ) humans run into medians and dividers every day.

Lets do some real comparisons here....

 

[Electroman]

I am supsecting that my AP2 car would have stopped in this situation, and in fact would have reduced speed *even* before the car in front of me darted away and exposing me to a stalled car.

Someone should this test it out.

Regarding the video of a Tesla slowing down for a crossing goose, that only happened because the bird was moving. If you put a bird shaped fixed object, I am pretty sure it won't

 

[PoitNarf]

Wait wait....hold on.

The car DOES NOT ignore stationary object in front of it. Lets get that clear.

The statement I was responding to was this: The car does not recognize stationary objects.

Conclusion: It does recognize stationary objects.

Now...my radar/camera based NoA has been flawless. It recognizes both moving and stationary cars in front of me daily. AND it does it better than it did a month ago....which makes me wonder why articles and videos from a year ago are valid.

Sure the software is improving dramatically and faster than I had expected, and you are correct that a Tesla can detect stationary objects in front of it just fine most of the time. My comment was specific to the shortcomings of using radar for detecting objects in front of the car while traveling at high speed. The radar does ignore tons of stationary objects in your driving path. Radar isn’t able to get a very clear picture of the environment in front of you which is why most stationary objects it picks up when traveling above a certain speed are ignored.

Tesla has been increasing their reliance on data from the cameras to properly map out the environment around the car, but they still rely on the forward facing radar for a lot of this data collection as well. These sort of accidents will continue to happen until Tesla is able to get the software to a level where the analysis of the camera data is able to identify these situations.

 

[Garlan Garner]

Sure the software is improving dramatically and faster than I had expected, and you are correct that a Tesla can detect stationary objects in front of it just fine most of the time. My comment was specific to the shortcomings of using radar for detecting objects in front of the car while traveling at high speed. The radar does ignore tons of stationary objects in your driving path. Radar isn’t able to get a very clear picture of the environment in front of you which is why most stationary objects it picks up when traveling above a certain speed are ignored.

Tesla has been increasing their reliance on data from the cameras to properly map out the environment around the car, but they still rely on the forward facing radar for a lot of this data collection as well. These sort of accidents will continue to happen until Tesla is able to get the software to a level where the analysis of the camera data is able to identify these situations.

Look.

I am making comments about radar/camera. Not just radar.

If Tesla's only had radar….NoA would be doomed.

Here....watch this.

This video is real. This video no doubt uses both radar and video at the same time.

The point is....there is no need for Lidar when you have object recognition like this. Watch the numbers on the "stationary" poles and objects in the distance.

Quiz as you watch video #1. What is a number 1,2,3 or 4 object? See if you can figure it out.
Quiz as you watch video #1. What does the yellow/pink/red/orange/green superimposed lane lines mean?

Hone in at 1:00 in the upper right of the screen and watch that bird get assigned an object number of 1 while its flying. Are you kidding me?

Watch both videos carefully and watch the green on the ground. Green means "drivable area". Watch as the drivable area changes and bends around objects and people as then enter and exit the "drivable area".

Look at the number of the stationary recognized simultaneous objects being recognized. Its astounding.

Does Tesla really need anything more than this?

After watching these videos.... are we still willing to say that Radar/Camera would run into a parked car? - and that an expensive LIDAR system is needed? really?

 

[davedavedave]

“Driver’s in the United States get into these kinds of accidents all the time... sometimes crisid is averted by the skin of their teeth.. but autopilot and Tesla both suck... boo Tesla, boo!!”

— Forbes Mag

Sorry.. the article is terrible. Accident or not.

I couldn't agree more!

"...Tesla ought to be using LIDAR, a mash-up of light and radar that is a sensory device..." - huh? No, it's not.

"There is also a chance that the human driver might have assumed that the Autopilot was going to aid in the driving...". No, not according to the article he sourced, and presumably read.

"Tesla typically points out in these kinds of incidents that it is the human driver that is responsible for the car, no matter what the Autopilot automation does or doesn’t do.". Of course. It's not legal to let a car drive unsupervised yet.

"...turns out that you were supposed to be the one handling the controls (a Catch-22, as it were).". No catch 22 here, just speciously inverted logic.

"I am Dr. Lance B. Eliot, a world-renowned expert on Artificial Intelligence (AI) and Machine Learning (ML)....". Declaring one's self to be "world-renowned" is unusual, unless you're a wrapper or a pro wrestler.

I wish there was a feature to block specific authors, similar to how one can block people on Facebook.

 

[PoitNarf]

Look.

I am making comments about radar/camera. Not just radar.

If Tesla's only had radar….NoA would be doomed.

Here....watch this.

This video is real. This video no doubt uses both radar and video.

The point is....there is no need for Lidar when you have object recognition like this. Watch the numbers on the "stationary" poles and objects in the distance.

Look at the number of the stationary recognized simultaneous objects being recognized. Its astounding.

Does Tesla really need anything more than this?

Yes I’ve seen these videos many times before and it’s crazy how sophisticated the current system is. However, what none of us know is the weights assigned to each input. Does Tesla assign greater weight to vision data or radar data? Maybe equal weights? Perhaps the weights are variable depending on the current driving conditions? Having video of what the sensor suite is detecting only gives us part of the picture. We have no idea how the AP computer is deciding on how to act on this data or which sensor data it thinks has the clearest picture of the environment in your driving path.

I have zero doubt that Tesla will achieve FSD with their current sensor suite once the software matures enough. None of us here are defending LIDAR. I’m all in with AP, NoA, FSD, etc. I understand the current limitations of the system and I was pointing out one of the shortcomings of a specific sensor in the sensor suite. Just because my Model 3 is able to correctly identify stationary vehicles while stopped at an intersection doesn’t mean it will be able to detect and automatically brake for a stalled vehicle in front of me on the highway while I’m going 75mph.

 

[Garlan Garner]

Yes I’ve seen these videos many times before and it’s crazy how sophisticated the current system is. However, what none of us know is the weights assigned to each input. Does Tesla assign greater weight to vision data or radar data? Maybe equal weights? Perhaps the weights are variable depending on the current driving conditions? Having video of what the sensor suite is detecting only gives us part of the picture. We have no idea how the AP computer is deciding on how to act on this data or which sensor data it thinks has the clearest picture of the environment in your driving path.

I have zero doubt that Tesla will achieve FSD with their current sensor suite once the software matures enough. None of us here are defending LIDAR. I’m all in with AP, NoA, FSD, etc. I understand the current limitations of the system and I was pointing out one of the shortcomings of a specific sensor in the sensor suite. Just because my Model 3 is able to correctly identify stationary vehicles while stopped at an intersection doesn’t mean it will be able to detect and automatically brake for a stalled vehicle in front of me on the highway while I’m going 75mph.

I don't care about weights... I don't care how their radar/camera suite is doing it.

To me its a non-issue

The point is: it recognizes objects and even recognizes what kind of objects they are - and we all can see...tons of additional data. More data than any human can compute at the same time.

It recognizes if the objects are stationary or moving - just like I do when I drive. I don't assign weights to what my ears her or nose smells or eyes see or hands touch.

I uses all of my senses simultaneously without weighting anything. Nothing more is necessary.

 

[Garlan Garner]

Just because my Model 3 is able to correctly identify stationary vehicles while stopped at an intersection doesn’t mean it will be able to detect and automatically brake for a stalled vehicle in front of me on the highway while I’m going 75mph.

You have never had NoA come to a stop for stopped traffic on a highway while you were going 75 mph?

I have.

I355 in the Chicagoland area - northbound - happens every day during rush hour.

 

[davedavedave]

Every writer critical of Tesla 100% always has an agenda, don't they? If it wasn't for all these agenda-driven writers, it would be nothing but 100% I love Tesla articles all day!

I immediately thought there was something fishy about this guy, and some quick googling confirmed it. From his Forbes bio: " Author of over 30 books..." But only one book on Amazon? https://www.amazon.com/Lance-B.-Eliot/e/B001KCRU8M?ref=dbs_a_def_rwt_hsch_vu00_taft_p1_i0

 

[AKinDC]

The reason this article is so ridiculous is something that wasn't mentioned...without AP, the exact same accident would have occurred. There's no way the human driver would have avoided it.
If you're going to criticize AP, it should be over something that it messed up that a human wouldn't have.

 

[Garlan Garner]

The reason this article is so ridiculous is something that wasn't mentioned...without AP, the exact same accident would have occurred. There's no way the human driver would have avoided it.
If you're going to criticize AP, it should be over something that it messed up that a human wouldn't have.

Excellent point.

Thanks for the conversation everyone - I now have to get going to cut my grass with my non-autopilot ( ICE ) lawn mower.

I need one of these.

https://www.lowes.com/pd/husqvarna-...HxLaVt9gZA8ccUrupscy5dvS9-ySHjQcaAo-mEALw_wcB

Maybe it wouldn't be so expensive if it didn't have LIDAR....just kidding.

 

[PoitNarf]

I don't care about weights... I don't care how their radar/camera suite is doing it.

To me its a non-issue

The point is: it recognizes objects and even recognizes what kind of objects they are - and we all can see...tons of additional data. More data than any human can compute at the same time.

It recognizes if the objects are stationary or moving - just like I do when I drive. I don't assign weights to what my ears her or nose smells or eyes see or hands touch.

I uses all of my senses simultaneously without weighting anything. Nothing more is necessary.

I think you’re putting more confidence in the current software implementation of AP than is warranted. Would you be willing to test this out yourself with your own Model 3 and recreate a similar situation to see if your car would brake on its own in time to avoid a collision? I certainly wouldn’t.

 

[PoitNarf]

You have never had NoA come to a stop for stopped traffic on a highway while you were going 75 mph?

I have.

I355 in the Chicagoland area - northbound - happens every day during rush hour.

Yes, because the traffic in front of me is slowing down and not stopped completely.

I’ve never encountered a situation where there is a stationary object directly in front of me while traveling at highway speeds. I hope I never encounter this.