Featured Tesla Autopilot recall probed by safety regulator following new crashes

a decade & more ago before anyone had plugins, there was data showing Prius drivers are more educated then none Prius drivers - typically more likely to have at least completed undergraduate work. Don't remember who started the thread & I guess I'm too over educated to go back & look

.

 

Well, my say is the driver's IQ is inversely proportional to the size of their pickup and with some pickup being EV now, it will reduce the IQ of the EV crowd lol.

 

Last month, Tesla indicated that they reached one billion miles driven in FSD and using this graph,

https://pbs.twimg.com/media/GKb5Wj2WQAEc5nG?format=jpg&name=small

We can interpolate that between August 2022 and August 2023, about 150 millions miles were driven in FSD. According to this NHTSA report,

https://static.nhtsa.gov/odi/inv/2022/INCR-EA22002-14496.pdf

There were 60 FSD related accidents (so 1 accident per 2.5 million miles) and one death (so 0.67 deaths per 100 million miles). It's one death too many but it's still less than manual driving, which in 2021, there were 1.37 deaths per 100 million miles,

Yearly snapshot

Over twice as much as FSD and FSD has gotten way better since the release of V12 and lots more miles were racked in.

 

Beware of selection biases that tilt or invalidate such comparisons. E.g.if FSD doesn't work on all roads in all conditions, but forces manual control in those cases, then manual-control deaths in those cases that FSD won't handle should be excluded from the comparison. Different roads and weather conditions have very different crash and fatality rates.

I've never heard of Tesla releasing enough data to make such apples-to-apples comparisons.

 

It hasn't been mentioned for FSD, but Musk said that any accidents where Autopilot was deactivated within 30 seconds of the accident was still considered an Autopilot accident. It would be fair to assume the same for FSD reported accidents.

 

That doesn't come even remotely close to covering all-road, all-weather selection biases.

We already know that for traditional un-assisted drivers, different types of roads have significantly different fatality rates. We need to start seeing FSD fatality rates separated into those same categories.

Given enough data, it would be conceivable to find FSD vs human-only ratios come out different for different road types and weather conditions. Identifying such differences will be a clue that we are starting to get better data.

 

I would not put much credence in the Tesla reported statistics. For instance, the only crashes that were included in the report were those that triggered the air bags. According to police statistics only 18 % of the accidents reported to them had airbags deployed.

From https://static.nhtsa.gov/odi/inv/2022/INCR-EA22002-14496.pdf

From that I can only surmise that the true number of accidents and injuries are about 5 times higher than reported.

Acording to the NHTSA's research, over 76% of the accidents were in situations where the driver had plenty of time to respond to hazard alerts but they failed to take action in time. That included collisions with stopped first responders that had plenty of lights and flares deployed.

 

I agree, you shouldn't use FSD in a winter storm but Interestingly enough, summer time, where the weather is usually the nicest, is where most road death per 100 million miles driven happen.

Motor Vehicle - Crashes by Month - Injury Facts

As for road type, FSD can be engaged on any type of roads, including dirt roads as I've shown.

 

there'll always be a bit of voodoo Juju fatality injury in accident statistics. Older cars ? someone runs off the road, or into a tree while inebriated Etc. or even 2 uninsureds both cars flee (criminal history) Those will skew stats favorably against internet linked / self drive assist vehicles because the modern systems knows what you did & where you did it. Dumb cars - not so much. Automotive mulligan. We had a sliding minivan door get crunched along the side of the garage. No one was the wiser
.

 

Duno ... auto corporation's (& the rest) credibility is fragile when it comes to reputation. Once discovered the manufacturer takes a beating if they're keeping secrets of serious safety problems. Look how Lexus & Toyota got a black eye when they were trying to hide unintended acceleration. Or the Ford Explorer / Firestone rollover snafu. It's a matter of candidness now or get blasted later. Haven't corporations learn from history?
.

 

This is what Tesla say about how they capture crash data.

"this correlates to nearly any crash at about 12 mph (20 kph) or above, depending on the crash forces generated". It's anytime an "airbag or other active restraint deploys", so airbags and seatbelt pretensioners.

Not much injuries happens at speed below the threshold of airbag deployment I would venture.

As for death, they are reported no matter what and data shows FSD had less death related accident than manual driving.

But yeah, most Autopilot accidents are caused by inattentive drivers.

On my 98 km FSD drive last weekend, on a section of segregated regional road, there were two cars parked on the shoulder. I was as it should, on the right lane, but as we approached those two cars, FSD move the car to the left lane all by itself. I hope Tesla update the Autopilot code with the FSD one as it's so much better.

 

And when Tesla reported their crash data to the NHTSA, instead of reporting from 10 days after the order was given to manufacturers in June 2022, they included crash data as far as 2019, so in Tesla's case, they over-reported their crash data, and they got dinged for it.

NHTSA report: Tesla led in crashes, but data has gaps

 

I have switched off FSD numerous times, but it seems to have been much improved lately.

The problem with it now is mainly lack of some nice user settings. I seem to put it on moderate and use the accelerator when its not going close enough, and why oh why to I have to hit minimize lane changes each time. Still it is rather predictable where it will not work well, and these situations are less and less. It is not self driving. The driver still needs to be alert in these situations and prepare to take over. I agree with some that it has been over sold, but used properely it makes you a safer not more dangerous driver. Still this is a learning curve. I would prefer that I take over versus the phantom braking. I also liked it better when it used radar cruise control but they have been removing radar from the 3 and Y so I get why we no longer get to use vision and radar.

My brother recently got a cadilac lyric. When he tried the tesla on vacation he didn't like that he had to keep his hands on the wheel. It's something that you get used to. The lyric doesn't let the auto steer work in many situations that the tesla does well, and that means it won't compare.

 

GM and Ford geofence their Level 2+ systems to highways that they have better mapping of. Toyota also geofence the TJA in the Prius Prime.

 

I think people are forgetting what's actually the core here.

The NHTSA is looking at data to avoid situations where people are letting go of the wheel. To your point about the lyric - letting someone let go of the wheel for too long - that's the problem. The NHTSA wants to make sure that the reminder systems are sufficient and aren't easily ignored or loopholed by putting something on the wheel to trick the car.

They are only going after Tesla because of the amount of data on Tesla they have.

Also whether autopilot or FSD make the average driver safer is immaterial to them. Their concern is the ratio of accidents that could have been avoided with driver intervention to those that could not. If half of the accidents using these systems could have been avoided if a driver was paying attention you have to explore - why weren't they paying attention and could they have been given better reminders, or could the system have forced them to take over sooner?

 

Driving a 717 mi Friday, I turned FSD off to cut to a mini-market bathroom and got hit in the rear bumper half-way into the parking lot:

• FSD OFF -> accident, thankfully minor.

So what has been your hand-on experience?

Bob Wilson

 

Level 2+ systems can allow hands off operation under certain conditions. The manufacturers that have them use driver monitoring beyond sensing hands on the wheel. Toyota adds cameras to the Prius Prime steering wheel, and Cadillac was using driver cameras before Supercruise and Tesla. The warnings and car response to inattentive drivers could be more aggressive than what Tesla does.

From the OP article, "Also, NHTSA noted that Tesla disclosed that the recall fix “requires the owner to opt in and allows a driver to readily reverse it.”"

Tesla released a safety recall fix that can easily be subverted. That's the reason.

 

Better mapping means squat when there is an obstacle on the road. That's why you need an attentive driver.

Beside even without map, FSD navigated my private road at the cottage, which do not appear in Google Map and OpenStreetMap for part of it (marked in red below) and what does show up is not even at the right location!



Here is it driving this road by itself.

 

Nice drive. All on FSD?

Bob Wilson

 

Which is why these systems have more than a torque sensor in the steering wheel to monitor driver attentiveness. The geofencing only pertains to using the hands free function. Can't say how effective they are as a hands on the wheel system elsewhere.