Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Another fatal autopilot crash - China

This site may earn commission on affiliate links.
This is another example of a driver being unwittingly turned into a zombie by Zombie Mode, er, Autopilot.

AP Teslas have a 13x higher fatality rate per mile than non-AP Teslas. One can claim that the driver wasn't paying enough attention and thus deserved to die, and that Tesla is therefore not responsible for their actions, but I am not interested in that argument.

Personally, I am interested in reducing fatalities by whatever means necessary. If Autopilot was never released until the features in 8.0 were present, two more people would be alive now, as 8.0 contains features that mitigate the original design flaws in AP. These flaws were known before release; they're even listed in the owners manual. Two foreseeable, preventable deaths from releasing early. AP Teslas produce 13x more fatalities per use than non-AP Teslas. That's the bottom line.

You could argue that releasing a flawed product early allows Tesla to get level 4 more quickly, this saving more lives in total and justifying the sacrifice. However, because AP is capable of operating in a completely passive mode, I strongly doubt lack of active control of cars would have slowed AP development and delayed 8.0.
 
Last edited:
This is another example of a driver being unwittingly turned into a zombie by Zombie Mode, er, Autopilot.

AP Teslas have a 13x higher fatality rate per mile than non-AP Teslas. One can claim that the driver wasn't paying enough attention and thus deserved to die, and that Tesla is therefore not responsible for their actions, but I am not interested in that argument.

Personally, I am interested in reducing fatalities by whatever means necessary. If Autopilot was never released until the features in 8.0 were present, two more people would be alive now, as 8.0 contains features that mitigate the original design flaws in AP. These flaws were known before release; they're even listed in the owners manual. Two foreseeable deaths from releasing early. AP Teslas produce 13x more fatalities per use than non-AP Teslas. That's the bottom line.

You could argue that releasing a flawed product early allows Tesla to get level 4 more quickly, this saving more lives in total and justifying the sacrifice. However, because AP is capable of operating in a completely passive mode, I strongly doubt lack of active control of cars would have slowed AP development and delayed 8.0.

What a load of pseudo-statistical crap.
 
This is another example of a driver being unwittingly turned into a zombie by Zombie Mode, er, Autopilot.

AP Teslas have a 13x higher fatality rate per mile than non-AP Teslas. One can claim that the driver wasn't paying enough attention and thus deserved to die, and that Tesla is therefore not responsible for their actions, but I am not interested in that argument.

Personally, I am interested in reducing fatalities by whatever means necessary. If Autopilot was never released until the features in 8.0 were present, two more people would be alive now, as 8.0 contains features that mitigate the original design flaws in AP. These flaws were known before release; they're even listed in the owners manual. Two foreseeable deaths from releasing early. AP Teslas produce 13x more fatalities per use than non-AP Teslas. That's the bottom line.

You could argue that releasing a flawed product early allows Tesla to get level 4 more quickly, this saving more lives in total and justifying the sacrifice. However, because AP is capable of operating in a completely passive mode, I strongly doubt lack of active control of cars would have slowed AP development and delayed 8.0.
Whoa! There is ONE confirmed fatality in a car with AP active...ONE! How does that get you to 13x higher than none AP Tesla's. Please provide a source for this data.
 
  • Helpful
Reactions: EarlyAdopter
...No technical fix can solve this...
The ONLY solution is to stop blaming the technology and start blaming the drivers.

I think with better hardware, there might not be those two fatalities.

The current limitations are:

1) Monochromatic, monocular camera that has no depth perception and it is blind sometimes such as failure to detect a big white tractor-trailer against brightly lit sky. A better one would be stereoscopic and all-color camera.

2) Missing LIDAR.

As a solution, Tesla improves on software instead of announcing more reliable hardware: V8 will prioritize radar over camera and it will do detailed point cloud just like LIDAR does.
 
Nature abhors a vacuum, and government always likes more rules:

Federal officials plan aggressive approach to driverless cars

I think that makes sense.

There needs to be a minimum standard that companies can show how well they pass or fail each scenarios so owners can take precautions.

Nvidia shows that it can navigate without hands on the steering wheel through traffic cones with 1 camera and 1 radar:

Ru6Sogn.png




Tesla has 1 camera, 1 radar and 12 ultrasonic sensors, so it is only fair that Tesla should show whether it can do the same without hitting a single cone:


https://teslamotorsclub.com/tmc/posts/1650041/

The highest legal speed limit in the US is 85 MPH in Texas, so the minimum standard should be that a car should be able to brake or steer around to avoid an obstacle on freeway with that speed.

Companies needs to publish their tests so owners know which tests they passed without hitting anything and which ones they still need to work on.
 
  • Like
Reactions: Zybd1201
I think that makes sense.

There needs to be a minimum standard that companies can show how well they pass or fail each scenarios so owners can take precautions.

Nvidia shows that it can navigate without hands on the steering wheel through traffic cones with 1 camera and 1 radar:

Ru6Sogn.png




Tesla has 1 camera, 1 radar and 12 ultrasonic sensors, so it is only fair that Tesla should show whether it can do the same without hitting a single cone:


https://teslamotorsclub.com/tmc/posts/1650041/

The highest legal speed limit in the US is 85 MPH in Texas, so the minimum standard should be that a car should be able to brake or steer around to avoid an obstacle on freeway with that speed.

Companies needs to publish their tests so owners know which tests they passed without hitting anything and which ones they still need to work on.

Tesla is not selling a driverless car. Nvidia is not ready for production.
 
  • Like
Reactions: green1
I think that makes sense.

There needs to be a minimum standard that companies can show how well they pass or fail each scenarios so owners can take precautions.

Nvidia shows that it can navigate without hands on the steering wheel through traffic cones with 1 camera and 1 radar:

Ru6Sogn.png




Tesla has 1 camera, 1 radar and 12 ultrasonic sensors, so it is only fair that Tesla should show whether it can do the same without hitting a single cone:


https://teslamotorsclub.com/tmc/posts/1650041/

The highest legal speed limit in the US is 85 MPH in Texas, so the minimum standard should be that a car should be able to brake or steer around to avoid an obstacle on freeway with that speed.

Companies needs to publish their tests so owners know which tests they passed without hitting anything and which ones they still need to work on.

Ever try to steer around an obstacle at 85mph?
 
Latest from Tuesday court, it looks like the family is now officially blocking Tesla from investigating the accident. That means no physical access for car log from Tesla personnel:

“The family insists the investigation should be done by a third party, rather than Tesla,”

Family of Driver Killed in Tesla Crash in China Seeks Court Investigation

Personally, I always figured this was the case. We were getting bits and pieces. Tesla said the family wouldn't let them investigate (by giving Tesla physical access themselves) and the family claimed the were (by telling Tesla, give us the tools and code to do it ourselves and we will provide you a copy). So with these "half-truths", we got "Tesla can't investigate b/c it won't release its proprietary information and the family is impeding Tesla's investigation b/c it won't give Tesla unfettered access to the vehicle".

I am NOT saying one side is right or wrong. But I'm not surprised that both statements of "They are not letting us investigate" and "Tesla can investigate but choosing not to" are true to a degree.
 
  • Like
Reactions: ohmman
Personally, I always figured this was the case. We were getting bits and pieces. Tesla said the family wouldn't let them investigate (by giving Tesla physical access themselves) and the family claimed the were (by telling Tesla, give us the tools and code to do it ourselves and we will provide you a copy). So with these "half-truths", we got "Tesla can't investigate b/c it won't release its proprietary information and the family is impeding Tesla's investigation b/c it won't give Tesla unfettered access to the vehicle".

I am NOT saying one side is right or wrong. But I'm not surprised that both statements of "They are not letting us investigate" and "Tesla can investigate but choosing not to" are true to a degree.
The part I agree with is involving an independent party for the research.
 
The part I agree with is involving an independent party for the research.
That's true, but expecting Tesla to give proprietary tools to a third party is pretty naive and beyond the scope of how investigations are typically done. The most they can ask for is Tesla cooperating with authorities and certifying under perjury certain facts are true (for example, saying whether autopilot was on or off is pretty straightforward). Even when the NHTSA here does investigations they rely on manufacturers to do a lot of the work.

The clarification makes things more clear.
 
  • Like
Reactions: Stoneymonster
That's true, but expecting Tesla to give proprietary tools to a third party is pretty naive and beyond the scope of how investigations are typically done. The most they can ask for is Tesla cooperating with authorities and certifying under perjury certain facts are true (for example, saying whether autopilot was on or off is pretty straightforward). Even when the NHTSA here does investigations they rely on manufacturers to do a lot of the work.

The clarification makes things more clear.
Typically the expert for the family would be in the room while Tesla makes a copy of the evidence and then analyzes it. We now also know, due to wk057 research, that there's likely video on the MCU. I know they already have the dashcam video so this would be additive.
 
That's true, but expecting Tesla to give proprietary tools to a third party is pretty naive and beyond the scope of how investigations are typically done. The most they can ask for is Tesla cooperating with authorities and certifying under perjury certain facts are true (for example, saying whether autopilot was on or off is pretty straightforward). Even when the NHTSA here does investigations they rely on manufacturers to do a lot of the work.

The clarification makes things more clear.

What exactly is so proprietary about accessing vehicle log data? Is that supposed to be some big secret? There is no reason at all for Tesla to refuse to an investigation by an independent third party. If someone wanted to access a Tesla all they would have to do is buy one. Nothing truly proprietary there.