”This Test Shows Why Tesla Autopilot Crashes Keep Happening”

[12Pack]

So would this have happened if the fake car had enough metal to register on the front radar?

https://jalopnik.com/this-test-shows-why-tesla-autopilot-crashes-keep-happen-1826810902

 

[thegruf]

this video is all over the internet, and it's f*ing irritating.
Even the BBC are touting this around as some sort of magical revelation.
The rest of the internet falling in linelike a bunch of iSheep

This video shows telsa AP performing exactly as it states it does in the manual.
Q How is this news?
A It isn't - it is pure sensationalism

Thatcham are supposedly a brand neutral research organisation funded by the uk insurance industry.

What Thatcham have done here though, is create a particular test condition with no published detail of the test, designed to demonstrate a known limitation in one manufacturers driving assistance suite even though said manufacturer clearly states in their manual accommpanying the product this is a circumstance the suite may not be reliably detect.

Now let's see the same test across a repesentative range of manufacturers products and see a genuine report rather than a selectively biased one that will inevitable impact on one manufacturer's perception in the marketplace

/rant

 

[Ludalicious]

So would this have happened if the fake car had enough metal to register on the front radar?

https://jalopnik.com/this-test-shows-why-tesla-autopilot-crashes-keep-happen-1826810902

Little concerning. What if that was a kid on the road?

 

[thegruf]

RTFM:

pp77/78

Warning: Traffic-Aware Cruise Control cannot detect all objects and may not brake/decelerate for stationary vehicles or objects, especially in situations when you are driving over 80 km/h and in situations where a vehicle you are following moves out of your driving path and a stationary vehicle or object is in front of you. Always pay attention to the road ahead and stay prepared to take immediate corrective action. Depending on Traffic-Aware Cruise Control to avoid a collision can result in serious injury or death. In addition, Traffic-Aware Cruise Control may react to vehicles or objects that either do not exist or are not in the lane of travel, causing Model S to slow down unnecessarily or inappropriately.

THIS. IS. A DESIGN. PARAMETER. OF. THIS. CAR. PERIOD.

It doesn't fly either, is that also concerning?
(Elon absolutely isn't - probably working on this though)

 

[Ludalicious]

RTFM:

pp77/78

Warning: Traffic-Aware Cruise Control cannot detect all objects and may not brake/decelerate for stationary vehicles or objects, especially in situations when you are driving over 80 km/h and in situations where a vehicle you are following moves out of your driving path and a stationary vehicle or object is in front of you. Always pay attention to the road ahead and stay prepared to take immediate corrective action. Depending on Traffic-Aware Cruise Control to avoid a collision can result in serious injury or death. In addition, Traffic-Aware Cruise Control may react to vehicles or objects that either do not exist or are not in the lane of travel, causing Model S to slow down unnecessarily or inappropriately.

THIS. IS. A DESIGN. PARAMETER. OF. THIS. CAR. PERIOD.

It doesn't fly either, is that also concerning?
(Elon absolutely isn't - probably working on this though)

Settle down TURBO

 

[Brando]

Little concerning. What if that was a kid on the road?

Driving a two and half ton vehicles is dangerous, especially when/if people can get in the way. wake up

Why do US drivers kill 2-4 times more people than much of the rest of the industrial world ??
List of countries by traffic-related death rate - Wikipedia

You can arrange the above table by each column
[look for the triangle at the top of each column and click away
- let us know if you find something noteworthy]

 

[WannabeOwner]

Its the same for all cars with Adaptive Cruise Control, they won't avoid a stationary object that they did not see slowing and stopping. Nothing specific to Tesla here.

Why emergency braking systems sometimes hit parked cars and lane dividers

The bit I don't get is that the current hardware is going to be able to do FSD, which will need to include stop-for/avoid stationary objects. So what's going to change to make that possible?

That said, I think cars with advance driver features need to behave how the driver expects. If I take a 17 year old for their first attempt at driving a car I know all the things that they might do wrong, and can take over or prevent them. With AP I have absolutely no idea what it might choose to do; it jumps on the brakes for no reason. I assume it thinks the gap next to a juggernaut is too narrow, but to my eye it looks exactly the same as the past X-hundred similar ones. That happens maybe once a month. The drive-into-the-median type accident happens maybe once in 100 car ownerships; maybe once in 1,000 ownerships; and only if the driver is not concentrating. There's a very real risk that, after mile-upon-mile of faultless AP driving, the driver is highly likely to be fooled into thinking that they can do other things ... text even ... and that's eyes off the road for a prolonged period, with potentially deadly consequences.

I have a hand on the wheel 100% of the time, and my right foot hovering over accelerator (in case of phantom braking when there might be traffic behind me which isn't expecting it as nothing around me that would warrant it)

But I arrive hugely more refreshed on AP than I used to manual-driving, so I'm happy. But I worry about folk that become complacent, or think that hands-on-lap is going to allow them to react in time; maybe they will, but I ain't taking that risk.

 

[Ludalicious]

That said, I think cars with advance driver features need to behave how the driver expects. If I take a 17 year old for their first attempt at driving a car I know all the things that they might do wrong, and can take over or prevent them. With AP I have absolutely no idea what it might choose to do; it jumps on the brakes for no reason. I assume it thinks the gap next to a juggernaut is too narrow, but to my eye it looks exactly the same as the past X-hundred similar ones. That happens maybe once a month. The drive-into-the-median type accident happens maybe once in 100 car ownerships; maybe once in 1,000 ownerships; and only if the driver is not concentrating.

I was thinking the exact same thing. Unless there is a sensor installed that's not being used for some reason. But wouldn't that be the first sensor they would calibrate and make sure is in working condition.

Maybe that's why Elon is building tunnels. Tunnels with sensors that will communicate with vehicles and notify them of objects ahead.

 

[chillaban]

The bit I don't get is that the current hardware is going to be able to do FSD, which will need to include stop-for/avoid stationary objects. So what's going to change to make that possible?

More advanced software is key, but it's also not ready yet. Cars like AP1 had woefully underpowered MIPS CPU cores (similar to what's in 10 year old wireless routers) paired with an ASIC for accelerating a pre-trained neural network architecture.

AP2 uses general purpose GPUs combined with 64-bit ARM CPUs, either of which by itself is more powerful than AP1, which allows Tesla to redefine what the car can do in terms of software.

Right now it's kind of like looking at a regular calculator vs a high-end gaming PC running Windows Calculator and wondering how they can claim that gaming PC is capable of rendering games in realtime but the 10-key calculator cannot, despite both performing very similar functions.

(Well that might be a bit of an exaggeration. Tesla still has a lot to prove about their capability to deliver on their software promises as well as WHEN they'll be able to do so.)

 

[McRat]

Its the same for all cars with Adaptive Cruise Control, they won't avoid a stationary object that they did not see slowing and stopping. Nothing specific to Tesla here.

...

I have no idea why folk think this. They should have had other cars do the test too.
I know the GM system will fire off the warnings (not subtle) in such a situation, but I've never been in the mood to ignore it and see if it will actually hit the car in front of me.

This situation happens constantly at stop lights. A person will change lanes when there is an empty lane so they aren't behind a car, or the line is shorter in one lane. So you will experience it a lot. Freeways too. If the FAB/AEB did not work with ACC, you'd never get warnings and there would be a lot of reported accidents with the system.

This a low speed test because of the room is limited, and they probably don't want to throw people around or choke them.

Many cars are not metal in the rear. If it's a paper plate car, there no significant metal. But radar does not need metal to work. It works with baseballs just fine, and plastic cars too.

 

[chillaban]

I have no idea why folk think this. They should have had other cars do the test too.
I know the GM system will fire off the warnings (not subtle) in such a situation, but I've never been in the mood to ignore it and see if it will actually hit the car in front of me.

This situation happens constantly at stop lights. A person will change lanes when there is an empty lane so they aren't behind a car, or the line is shorter in one lane. So you will experience it a lot.

This a low speed test because of the room is limited, and they probably don't want to throw people around or choke them.

Many cars are not metal in the rear. If it's a paper plate car, their no significant metal. But radar does not need metal to work. It works with baseballs just fine, and plastic cars too.

The low-speed equivalent tests are pretty solid on Teslas too. I have not had a failure approaching a stopped cat at 55mph or below in probably 6 months now (ever since 2018.10.4). It's mainly that currently, the recognition of cars at all starts from 400-600ft (it's been increasing with recent updates). And the changing of ACC lead-car targets takes about 1-2 seconds (presumably a hysteresis to avoid rapid flip-flopping between targets).

In the BBC video there was all of a second or two of reaction time for the system to see and react to the stopped car because it was previously following a moving lead car.

 

[buttershrimp]

this video is all over the internet, and it's f*ing irritating.
Even the BBC are touting this around as some sort of magical revelation.
The rest of the internet falling in linelike a bunch of iSheep

This video shows telsa AP performing exactly as it states it does in the manual.
Q How is this news?
A It isn't - it is pure sensationalism

Thatcham are supposedly a brand neutral research organisation funded by the uk insurance industry.

What Thatcham have done here though, is create a particular test condition with no published detail of the test, designed to demonstrate a known limitation in one manufacturers driving assistance suite even though said manufacturer clearly states in their manual accommpanying the product this is a circumstance the suite may not be reliably detect.

Now let's see the same test across a repesentative range of manufacturers products and see a genuine report rather than a selectively biased one that will inevitable impact on one manufacturer's perception in the marketplace

/rant

That is some really bad television. I can't stand that condescending investigative and smug lesson from that wanker about "Autopilot", where's the part where they crash the Mercedes and BMW into the inflatable doll car? It's like false equivalency that tries to argue BMW and other car companies are in the same ballpark as Tesla. So dumb.

 

[MIT_S60]

That is some really bad television. I can't stand that condescending investigative and smug lesson from that wanker about "Autopilot", where's the part where they crash the Mercedes and BMW into the inflatable doll car? It's like false equivalency that tries to argue BMW and other car companies are in the same ballpark as Tesla. So dumb.

And then these videos only add to getting nags down to 15 second intervals...

 

[Twiglett]

Little concerning. What if that was a kid on the road?

Pretty obvious really - the kids dead and its your fault.

 

[McRat]

The low-speed equivalent tests are pretty solid on Teslas too. I have not had a failure approaching a stopped cat at 55mph or below in probably 6 months now (ever since 2018.10.4). It's mainly that currently, the recognition of cars at all starts from 400-600ft (it's been increasing with recent updates). And the changing of ACC lead-car targets takes about 1-2 seconds (presumably a hysteresis to avoid rapid flip-flopping between targets).

In the BBC video there was all of a second or two of reaction time for the system to see and react to the stopped car because it was previously following a moving lead car.

Yes, I hear this all the time. But I'm not willing to do a 70mph test against a real car. It does fire off the warnings at high speed. Seat vibrator, warning noise, 6 bright LEDs splayed across the windshield, icon on the HUD, whatever systems the car uses.

I've only logged about 24,000mi and 20 months with these systems, but I can say they ID threats like stopped cars, and some threats that are not documented.

Approaching an intersection at roughly 50 mph. My light is green, I'm in the #1 lane of a 4 lane divided road.
As I get pretty close, a Honda who was in the left turn bay who had a red light, suddenly turns right, right in front of me to make a right hand turn. It would have been a ~75° impact nearly broadside. My windshield lit up like fireworks and it made a racket. When?

undocumented behavior, cross traffic threat not directly in front of car, ACC off...

The instant the Honda's bumper crossed the solid white line defining the turn bay and was moving into my lane. The Honda was not actually a threat when it fired off. I would have cleared the idiot if he stopped instead of gassing it.

Did my car hit the brakes first or did I? That I cannot say. I'm normally about .22 seconds R/T consistently. There was no "out". It was brakes or nothing. The driver did a full right turn at a red lighted left turn bay at full throttle without looking. I hit the brakes with a straight wheel to ABS pulse, and avoided a collision. Without the system would I have hit him? I cannot say. I try to stay sharp. Would AP1/AP2 have hit the car? Give it a try.

 

[WannabeOwner]

where's the part where they crash the Mercedes and BMW into the inflatable doll car

Perhaps because the BBC has "previous" with Tesla over the Top Gear "legals"? ... are they heading for another trip to the Old Bailey by doing a one-brand-sided piece to camera ... again ... ?

Give it a try.

I have no idea why folk think this

I reckon because folk run into stationary objects ... now and again. If it can see the object some of the time that's great, all the time would be better, but seems like we aren't there yet (or perhaps there is some other reason folk run into stationary objects? given the manual says its not a guaranteed situation I assume there are known failure scenarios).

All the "Someone hit something" are making the click-bait news, you and I have both got "It avoided an impact" stories, but we didn't make the news. My HW1 car certain does the light-show and screaming-trick when it thinks I'm not doing a good enough job, soon enough. For example, I have had situation where I was approaching stationary traffic and left my braking later than the car would like such that it thought I hadn't noticed, so that was ipso facto in plenty of time for me to react and brake (given that I was just about to anyway)

 

[MrAustraliaTax]

I am really surprised at the negativity here on what is a very balanced and factual article.

The video showcases a known problem with AutoPilot. There are many, many posts here on TMC describing exactly this sort of crash and how it surprised the Tesla driver that the car did not respond as expected.

This video should be shown to EVERY Tesla driver before they leave the showroom. It will save lives and will prevent/mitigate many unnecessary crashes.

(Yes, this scenario is described in the manual. Yes, drivers should read the manual. But they don't.)

 

[Peteski]

The original Thatcham report was quite fair and was not aimed specifically at Tesla, it was merely to raise awareness of the limitations with ALL current semi-autonomous driving aids. It's the media reporting that has made it look like an anti-Tesla piece.

 

[Peteski]

I have no idea why folk think this. They should have had other cars do the test too.

They did actually test a BMW and Mercedes and they both failed this test too. From the in-car footage in another video the Tesla actually did fire off an AEB warning and then started braking, but all too late and we know that Tesla AEB only reduces speed by 25 mph anyway. I believe in this case it only managed to slow down by 5 mph before impact. An alert driver could have reacted to the AEB warning and swerved just like the car in front. A complacent driver not paying any attention would probably crash, which was the point of the report. Even a manually driven car with a distracted driver probably would have crashed too, but that's another debate all of its own!

I think the Thatcham report was fair and not specifically picking on Tesla. They did show the Tesla failing this particular test and they showed the BMW failing to follow road markings. They even stated how good the Tesla steering assist is compared to the others. But the gutter media decided just to focus on the Tesla parts because it gets more clicks.

 

[Peteski]

The bit I don't get is that the current hardware is going to be able to do FSD, which will need to include stop-for/avoid stationary objects. So what's going to change to make that possible?

Safe FSD is still a long way off IMO. Too many edge cases and unpredictable human drivers around for it to guarantee safety. AP2 is a million miles away from being FSD. It can follow simple lines and avoid hitting objects in simplistic scenarios, but that's as far as it goes. It struggles with stationary objects, it struggles to react to vehicles cutting in front of you at the last moment, etc, etc, etc. It's just a fancy cruise control with lane assist and the rest is a pipe dream for the foreseeable future.