NHTSA Close to forcing recall of FSD after more crashes

[Smokin]

EVERY auto manufacturer peddling some level of driver assistance in vehicles has room for improvement. That said, from the latest from Mercedes Benz and Audi to GM, no one’s driver assistance is as good as Tesla’s. I can say this because I have owned all of them. Maybe those are just so horrifically unusable that many don’t rely on them. So while we are all speculating why aren’t more questioning the motives of the NHTSA? The political motives of the NHTSA are suspect. Pete Buttigieg is a well known anti Tesla politician who now has influence over the NHTSA. He is also pro union and insofar it’s not a far cry to see that autonomous driving technology is a threat to big labor in trucking and transportation. This administration has made edicts against automation of ports amidst this supply chain crisis. Automation is a threat to forced wage increases. Elon has also been fairly critical of this administration who has already weaponized nearly every agency against its critics. If you’re not going “hmmmm” then you’re not objective.

 

[Supcom]

Without a method of recognition without some sort of radar FSD will not function 100%. What happens to Telsa cars when the weather changes during an unmanned drive and a message comes up vision not working. Who takes over the wheel? Do you think the company that has gone to outer space does not know this? While a great tool and I love it FSD will as configured now will never work under all conditions.

Radars/Lidars cannot read traffic lights, so all AVs will needs to slow down or pull over in the event of heavy enough rain. If there is no wheel to take over, then the car must wait.

 

[texas_star_TM3]

anyone surprised? Driving from Dallas to Taos NM on the stretch to Amarillo (flat and long straight roads) I experienced *TWO* fun phantom braking events ... going from ~80mph down to 60mph within seconds for no apparent reason is "fun". And I was using cruise control only. Never experienced phantom braking to this extent with legacy OEM "dumb" lane keep + cruise control assist.

 

[texas_star_TM3]

Yep.

They already got their out by redefining it as an L2 system.

the L2 part - not even certified as hands-free - is what gets me. So it's "FSD" but if you don't keep your hands on the wheel (at all times) and crash ... it's your fault. Meanwhile Mercedes is certified as L3 (although at quite low speeds) in Europe - but you can legally read a newspaper or watch a movie and the manufacturer takes on responsibly. BlueCruise and Super Cruise at least allow you to take the hands off the wheel and just look at the road.

 

[Dewg]

Ugh... Come on people, the SAE level classification is purely for regulatory compliance and liability during testing. Initial goals were L4, but to protect the company during the tough time of transitioning to Tesla Vision and training the neural nets, Tesla dropped SAE to L2. The decision to move to vision only is controversial, but something Musk feels passionate about. He may end up failing in the end, and ahead of his time on computer processing, but he's giving it a go. If he succeeds, the SAE level will pop back to L4, and he'll disrupt the market. If he fails, it'll stay L2, and no doubt cause financial problems for the company. But in the end, he took a risk and tried it. Progress is sometimes messy and risky, but the ends may also justify the means.

 

[DayTrippin]

Ugh... Come on people, the SAE level classification is purely for regulatory compliance and liability during testing. Initial goals were L4, but to protect the company during the tough time of transitioning to Tesla Vision and training the neural nets, Tesla dropped SAE to L2. The decision to move to vision only is controversial, but something Musk feels passionate about. He may end up failing in the end, and ahead of his time on computer processing, but he's giving it a go. If he succeeds, the SAE level will pop back to L4, and he'll disrupt the market. If he fails, it'll stay L2, and no doubt cause financial problems for the company. But in the end, he took a risk and tried it. Progress is sometimes messy and risky, but the ends may also justify the means.

The only reason I think Elon gave it a go was because logistics likely impacted his ability to source radars thus impacting his delivery numbers and ultimately his bonus. I had one of the first Y's to come without radar and TACC/AP was a total cluster truck. If I had known had bad it was going to be, I would have either not bought my Y or punted delivery wayyyyyyyyyyyyyyy down the road. It was a total *sugar* show and they should have waited to make the move to full Teslavision when the drivability was at least somewhat comparable between the cars that had radar and those that didn't.

Based on my personal experience, it took almost a year to get it close. I can also compare my refreshed S w/radar vs my '22 Model 3 without, and the S has had no PB events and the 3 a few. In general, my S seems to sense following distance better and is smoother on braking and acceleration with TACC/AP than my 3.

Don't know if that is all due to TACC/AP differences but I can tell you that a moderate throttle input on my S in insane results in FAR more acceleration than my 3. So they did a good job managing the throttle inputs on the S given all the power it has. Not that the 3 is bad, but by comparison to my S, it is. The 3 is probably better than a lot of cars I've driven with respect to the smoothness of throttle inputs.

Elon sold a bunch of horse manure when they yanked the radar and tried to sell it as an upgrade. Tesla lost a ton of my trust and goodwill as a result. Also means when I replace the 3 or S, I'll definitely look outside the Tesla ecosystem a lot more actively than I might have before.

 

[texas_star_TM3]

Ugh... Come on people, the SAE level classification is purely for regulatory compliance and liability during testing. Initial goals were L4, but to protect the company during the tough time of transitioning to Tesla Vision and training the neural nets, Tesla dropped SAE to L2. The decision to move to vision only is controversial, but something Musk feels passionate about. He may end up failing in the end, and ahead of his time on computer processing, but he's giving it a go. If he succeeds, the SAE level will pop back to L4, and he'll disrupt the market. If he fails, it'll stay L2, and no doubt cause financial problems for the company. But in the end, he took a risk and tried it. Progress is sometimes messy and risky, but the ends may also justify the means.

lol. Tesla was *never* L4 - let alone L3. They always (!) said to keep hands on the wheel and eyes on the road. That by definition isn't even L3. And at the end of the day it's up to the regulators (and not Elon) what Level their system ends up being certified.

 

[DJT1]

Correct - But towers watch planes in the air with radar, don't they? How does that work with rain?

Aircraft have transponders. When ATC queries a plane via a VHF coded transmission the transponder picks it up and then sends back a response, via VHF radio, and that is then shown on the controllers radar. There is a 4 digit code put into the aircrafts transponder that is given to us by ATC so they can tell airplanes apart. The transponders aren't affected by rain or weather.

Now, the controller can switch their radar to show all targets (even those without transponders) but they do so rarely and only if they're trying to find a plane that has lost its transponder due to failure. That type of radar is affected by rain/ice its similar to the airplanes weather radar. Picks up moisture and other things that reflect radio waves like mountains or metal aircraft.

A few years ago Malaysian Flight 370, the 777 that went missing (and is still missing), the pilot deliberately turned off the airplanes transponder so controllers couldn't see it anymore, and it changed direction. Several other radar installations (military) were using the primary, or all target mode, radar and did see it flying in a different direction until it flew far enough away that even that radar couldn't see it anymore because of the curvature of the Earth.

 

[Dewg]

Aircraft have transponders. When ATC queries a plane via a VHF coded transmission the transponder picks it up and then sends back a response and that is then shown on the controllers radar. There is a 4 digit code put into the aircrafts transponder that is given to us by ATC so they can tell airplanes apart. The transponders aren't affected by rain or weather.

Now, the controller can switch their radar to show all targets (even those without transponders) but they do so rarely and only if they're trying to find a plane that has lost its transponder due to failure. That type of radar is affected by rain/ice its similar to the airplanes weather radar. Picks up moisture and other things that reflect radio waves like mountains or metal aircraft.

Hense RADAR in cars will not work well in rain/snow, etc.

 

[Dewg]

Based on my personal experience, it took almost a year to get it close. I can also compare my refreshed S w/radar vs my '22 Model 3 without, and the S has had no PB events and the 3 a few. In general, my S seems to sense following distance better and is smoother on braking and acceleration with TACC/AP than my 3.

I got one of the last Model Y's with RADAR (early 2021), and my PBs were quite bad, especially near overpasses and large freeway signs. It's gotten much better over time, and recently very minor with vision only on FSD Beta. I'm not sure about following distance and acceleration. The visualizations seem to accurately gauge distance to the car in front, even though it's vision only. When in stop and go traffic, the car in front moves pretty realistically as they pull away. I think the jerky starts/stops is a calibration of the system. Previous versions were smoother, but some people complained they didn't accelerate fast enough and were pissing off people behind them. It's likely just a matter of tuning for the goldilocks zone.

 

[Supcom]

Hense RADAR in cars will not work well in rain/snow, etc.

Rain impact on radar is high dependent on frequency, especially when it's above 20 GHz. Aircraft tracking radars typically are around 2.7 GHz and transponders operate at about 1 GHz. Auto radars are typically well above 20 GHz.

BTW weather radar frequencies are chosen so that they ARE affected by rain.

 

[Dewg]

Rain impact on radar is high dependent on frequency, especially when it's above 20 GHz. Aircraft tracking radars typically are around 2.7 GHz and transponders operate at about 1 GHz. Auto radars are typically well above 20 GHz.

BTW weather radar frequencies are chosen so that they ARE affected by rain.

Interesting article:

How can self-driving cars 'see' in the rain, snow and fog?

A team of electrical engineers at University of California, San Diego is working on technology to help self-driving cars navigate safely in inclement weather.

www.abc10.com

 

[texas_star_TM3]

Hense RADAR in cars will not work well in rain/snow, etc.

but vision works great in dense rain / snow / darkness / fog / blinded by low sun ?

 

[Dewg]

but vision works great in dense rain / snow / darkness / fog / blinded by low sun ?

Nope, but neither do human eyes.

 

[AndrewZ]

Apparently the NHTSA is gearing up to release data collected from manufacturers after issuing their L2 ADAS reporting standing order last year -- it sounds like numbers provided to the press in advance do not bode well for Tesla

I scooped this out of an article that may be paywalled for you, but you can find it via Google and I imagine more articles will pop up

The government will soon release data on collisions involving vehicles with autonomous or partially automated driving systems that will likely single out Tesla for a disproportionately high number of such crashes.

In coming days, the National Highway Traffic Safety Administration plans to issue figures it has been gathering for nearly a year. The agency said in a separate report last week that it had documented more than 200 crashes involving Teslas that were using Autopilot, “Full Self-Driving,” Traffic-Aware Cruise Control or some other of the company’s partially automated systems.

Tesla’s figure and its crash rate per 1,000 vehicles was substantially higher than the corresponding numbers for other automakers that provided such data to The Associated Press ahead of NHTSA’s release. The number of Tesla collisions was revealed as part of a NHTSA investigation of Teslas on Autopilot that had crashed into emergency and other vehicles stopped along roadways.

Tesla does have many more vehicles with partly automated systems operating on U.S. roads than most other automakers do — roughly 830,000, dating to the 2014 model year. And it collects real-time data online from vehicles, so it has a much faster reporting system. Other automakers, by contrast, must wait for reports to arrive from the field and sometimes don’t learn about crashes for months.

In a June 2021 order, NHTSA told more than 100 automakers and automated vehicle tech companies to report serious crashes within one day of learning about them and to disclose less-serious crashes by the 15th day of the following month. The agency is assessing how the systems perform, whether they endanger public safety and whether new regulations may be needed.

General Motors said it reported three crashes while its “Super Cruise” or other partially automated systems were in use. The company said it has sold more than 34,000 vehicles with Super Cruise since its debut in 2017.

Nissan, with over 560,000 vehicles on the road using its “ProPilot Assist,” didn’t have to report any crashes, the company said.
Stellantis, formerly Fiat Chrysler, said it reported two crashes involving its systems. Ford reported zero involving its “Blue Cruise” driver-assist system which went on sale in the spring, though Ford wouldn’t say if there were crashes with less-capable systems.
GM said the three crashes weren’t the fault of Super Cruise. It also reported two crashes that happened before the June 2021 order, a spokesman said.
Several automakers and tech companies, including Toyota and Honda, declined to release their numbers before the NHTSA data is revealed.

A message was left seeking comment from Tesla, which has disbanded its media relations department. NHTSA wouldn’t comment on the data Tuesday.
Raj Rajkumar, an electrical and computer engineering professor at Carnegie Mellon University who studies automated vehicles, said he wouldn’t be surprised if Tesla was found to have had a high number of crashes involving its driver-assist systems. Tesla, based in Austin, Texas, stopped using radar in its system and instead relies solely on cameras and computers — a system that Rajkumar calls “inherently unsafe.”

The system’s computer, he said, can recognize only what’s in its memory. Flashing lights on an emergency vehicle, Rajkumar said, might confuse the system, as would anything that the computer hasn’t seen before.

“Emergency vehicles may look very different from all the data that the Tesla software had been trained on,” he said.
In addition to the publicly released crash data, NHTSA has sent investigative teams to far more incidents involving Teslas using electronic systems than other automakers. As part of a larger inquiry into crashes involving advanced driver assistance systems, the agency has sent teams to 34 crashes since 2016 in which the systems were thought to have been in use. Of the 34 crashes, 28 involved Teslas, according to a NHTSA document.

NHTSA said in documents that it has received 191 reports of crashes involving Teslas on Autopilot and nonemergency vehicles, plus 16 more involving parked emergency vehicles or those with warning lights, for a total of 207. Of the 191, the agency removed 85 because of actions of other vehicles or insufficient data to make a firm assessment of the crashes. That left 106 that were included in the Autopilot investigation.

It wasn’t clear if 207 matched the total number of Tesla crashes reported to NHTSA under the order. A NHTSA spokeswoman wouldn’t comment.
The agency ordered automakers and tech companies to report crashes involving driver-assist systems, as well as fully autonomous driving systems.
In defending its partially automated systems, Tesla has said that Autopilot and “Full Self-Driving” cannot drive themselves, and that drivers should be ready to intervene at all times. The systems can keep cars in their lanes and away from other vehicles and objects. But in documents released last week, NHTSA raised questions about whether human drivers can intervene fast enough to prevent crashes.

Tesla’s “Full Self-Driving” is designed to complete a route on its own with human supervision, with the eventual aim of driving itself and running a fleet of autonomous robo-taxis. In 2019, Musk had pledged to have the robo-taxis running in 2020.

Tesla’s Autopilot driver-assist system detects hands on the steering wheel to make sure drivers are paying attention. But that’s inadequate, Rajkumar said. By contrast, systems such as GM’s monitor a driver’s eyes with a camera, he said, to make sure they’re looking forward.

For people who are watching this closely and have been waiting for good data, I imagine this will have some satisfying levels of granularity for comparison

 

[Dewg]

Apparently the NHTSA is gearing up to release data collected from manufacturers after issuing their L2 ADAS reporting standing order last year -- it sounds like numbers provided to the press in advance do not bode well for Tesla

I scooped this out of an article that may be paywalled for you, but you can find it via Google and I imagine more articles will pop up



For people who are watching this closely and have been waiting for good data, I imagine this will have some satisfying levels of granularity for comparison

One thing I've noticed, and I'm not sure why this is the case, there are lots of videos of people (usually younger) posting videos of them breaking safety protocols in Tesla's vs other manufacturers. You don't see kids trying to break other adaptive cruise control with lane assist features and jumping into the back seat while their cars drive.

What is it about Tesla that gets more attention for that?

 

[DJT1]

Hense RADAR in cars will not work well in rain/snow, etc.

Right, Radar whether Civilian or not is not going to see very well through rain/snow it will bounce off and back to the transmitter and will show distance to the precipitation. If you want to see through rain/snow, to some degree, you would need multiple millimeter radar transducers that can interpolate between their two or more viewpoints to see the rain from different angles and spot holes in the rain and therefore see beyond it. If it is just fog, radar can see through it quite well.

Vision based solutions like visible spectrum cameras and Lidar aren't going to be able to see through dense rain or fog. IR cameras might resolve more in certain cases.

Now if nearby cars could share sensor data with each other, that would be a giant leap forward.

 

[Mardak]

I imagine this will have some satisfying levels of granularity for comparison

Seems like the data reported here was crashes with the feature active and total vehicles with feature available. I wonder if NHTSA also requested mileage that the feature is active or even crashes avoided, but how many companies could even report those numbers especially if they would otherwise only get data back from reported crashes?

Hopefully there won't be some conclusion along the lines of counting the number of crashes where Automatic Emergency Braking was available and deciding that almost all had AEB, so the feature should be recalled whereas there probably were a lot more crashes avoided because of it.

 

[AndrewZ]

One thing I've noticed, and I'm not sure why this is the case, there are lots of videos of people (usually younger) posting videos of them breaking safety protocols in Tesla's vs other manufacturers. You don't see kids trying to break other adaptive cruise control with lane assist features and jumping into the back seat while their cars drive.

What is it about Tesla that gets more attention for that?

My personal belief is honestly that it can likely be chalked up to early messaging from Elon, naming of the systems, the big promises made and that are still being made, and maybe Tesla’s appeal to younger people willing to take more risks, then combine all that with lacking driver engagement systems that can be circumvented by a citrus fruit placed in the steering wheel spokes.

Likely almost nobody believes a Nissan with ProPilot can drive itself, but there are people out there right now who still genuinely believe that Autopilot can drive itself to the point where the person behind the wheel can stop paying attention. I even see people selling used Teslas advertising them as self-driving in no uncertain terms. No not advertising FSD functionality or AP functionality, the tag line says “BUY THIS SELF-DRIVING TESLA“ or something to that effect.

Seems like the data reported here was crashes with the feature active and total vehicles with feature available. I wonder if NHTSA also requested mileage that the feature is active or even crashes avoided, but how many companies could even report those numbers especially if they would otherwise only get data back from reported crashes?

Hopefully there won't be some conclusion along the lines of counting the number of crashes where Automatic Emergency Braking was available and deciding that almost all had AEB, so the feature should be recalled whereas there probably were a lot more crashes avoided because of it.

This is an example of the data request letters sent out last year detailing what was expected, same letter sent to many manufacturers with the header changed

https://static.nhtsa.gov/odi/inv/2021/INIM-PE21020-85284.pdf

 

[texas_star_TM3]

One thing I've noticed, and I'm not sure why this is the case, there are lots of videos of people (usually younger) posting videos of them breaking safety protocols in Tesla's vs other manufacturers. You don't see kids trying to break other adaptive cruise control with lane assist features and jumping into the back seat while their cars drive.

What is it about Tesla that gets more attention for that?

other manufacturers don't call their lane-keep assist L2 system "auto pilot" and heavily promote "FSD" to be ready just any moment now...