Tests show all driver assist systems can be fooled
jalopnik.com
jalopnik.com
But declaring Level 2 as garbage is stretching it. I am driving Level 2 (ACC plus lane assists plus speed sings recognition plus a few other things) for a few years now and I definitely get less tired on long drives. I recently drove coast to coast and it was more boring than tired because of the Level 2.
When I hear reports of accidents in Teslas because the driver "took a nap" or watched a movie or whatever, it's hard to feel empathy, because that's just reckless driving at this level.
Of course Tesla is to blame when they marketed, and maybe are still marketing, level 2 capabilities as being on par with an "Auto Pilot".
My car also requires me to put actual pressure on the wheel at all times, I think a 10-15 second lapse is allowed, then it starts to beep.
We do have the anecdotal stories of people buying RVs with "Cruise control", enabling it on a freeway and going to the back to make a cup of coffee. Then being all surprised_pikachu.jpg when they're being dug out from the wreckage.
I also don't "feel" like it can drive itself so paying attention still is warranted.
Rather than saying Level 2 is garbage, we should look at the stats and determine whether it is or is not safer than driving without automation. A perfectly attentive driver will beat automation, but as all the annual deaths while driving show, that is certainly not how universally drive.
Level 2 is not intended to be autopilot. It's intended to be safer than driver unassisted.
Then Tesla should change the name and offer refunds to anyone who bought Full Self Driving.
Arguably, the incentives today aren't strong enough given the level of pedestrian fatalities
So I believe there's a tradeoff here. People drive to work all the time staring at their phones and only glancing up at the road every few seconds. They always have the option of turning the assist systems off and going back to distracted driving the old fashioned way. Obviously companies have no financial incentive to do this, but I posit that not making it impossible to use assist systems while completely distracted is actually a net benefit for road safety.
I'd rather have Honda Sense driving a car down the interstate by itself than some dude who's only looking up when the road curves.
Do you know people who drive like this? Driving this way is (a) very illegal and (b) recklessly endangering other people. I doubt it's especially common, at least in most of the US.
Reporter: "Sir, there's a picture that went out on Twitter this morning of you reading while still driving on the Gardiner [Expressway]."
Ford: "Yeah, probably. I'm busy."
Reporter: "So you read while driving?"
Ford: "Yeah, probably, yeah. I'm try[ing] to catch up on my work and you know I keep my eyes on the road, but I'm a busy man."
In October, Ford was accused of illegally dialing numbers on his cellphone and talking on it as he steered his gold minivan westbound along Dundas Street West near Spadina Avenue.
> Fisher next placed a small, weighted chain on the steering wheel, to simulate the weight of a driver’s hand, and slid over into the front passenger seat without opening any of the vehicle’s doors, because that would disengage Autopilot. Using the same steering wheel dial, which controls multiple functions in addition to Autopilot’s speed, Fisher reached over and was able to accelerate the vehicle from a full stop.
Yes, the good old easy method of putting a weighted chain to the steering wheel and climbing over to the passenger seat.
The problem with building idiot-proof things is that idiots are way too inventive.
https://www.caranddriver.com/news/a37260363/driver-assist-sy...
It costs hardware to run and cost to develop/test, and car manufacturing is all about price.
But also:
Today the algorithm knows the 3D geometry of your head, and knows that it does not makes sense with a flat face. But it is more important to get it to work good on everyone than to catch foolers.
Disclaimer: i'm one of all the developers behind camera monitoring.
"tests show human drivers can be fooled too"
A driver nag is an attentiveness reminder. You're driving a car. You're expected to be paying attention. No one else is responsible for safety. So the car wants (for both safety and liability reasons) to make sure you're doing what you're supposed to do. And it works. The car nags me and I tug the wheel to prove I'm there. Most of the time I'm paying attention anyway, but sometimes it's because I got distracted. It works.
If a driver deliberately wants to behave unsafely, there's no way to stop that, because there's an infinite number of ways drivers can behave unsafely. Do we want a system to detect when drivers speed up on yellow? Tesla could absolutely implement that. Why aren't we demanding it? Prevent drivers from approaching cars in adjacent lanes with higher than a 30mph speed delta? Could do that too. Prevent accelerating in the wrong direction on a road? Possible.
But no, all that stuff would be dumb, because it doesn't correspond to an actual safety problem with the hardware. We recognize it's driver behavior at fault, and we don't want our cars regulating behavior.
Why is this different?
I submit it's different because "Actually Elon Musk is an Asshole and TSLA went to the moon". Which isn't a very good premise for automobile safety designs.
It's because the premise is so weak. I can put a brick on the gas pedal, so no car is safe!
Is there any evidence that drivers are abusing Autopilot etc. any more than cruise control and causing a higher rate of accidents due to inattentiveness?
I was especially disappointed in Consumer Reports, which insinuated that these methods of "fooling" Autopilot were somehow connected to a fatal crash in Texas. https://www.consumerreports.org/autonomous-driving/cr-engine...
You have to get 11 paragraphs in before you get to this:
> Our evaluation does not provide specific insight into the Texas crash, but safety advocates and researchers at CR say that it does show that driver monitoring systems need to work harder to keep drivers from using the systems in foreseeably dangerous ways.
By this standard, cruise control should never have launched in any vehicle until technology was capable of verifying driver attentiveness with eye tracking.
Faking a "malfunction" and then suing could be part of the threat model to the car maker and/or the insurance companies that provide liability coverage.
Because the big shiny text that people pay $10,000 for says "Autopilot". Between the name and the large price tag, it conveys that it's higher than L2 autonomy. Putting in the fine print that it's not, doesn't change things.
But I do, I did, and it does. So there.
asking becuase the message yours responds to says exactly that: people whoy actually buy autopilot do know very well that it is a system that requests and sometimes even requires driver's attention.
so if you confirm that, then how is a meta argument, ridiculously so?
People who understand how English works? Like, if I spend 10k on something that is "Full Self Driving", I'm paying for the ability for it to drive itself, fully. That is to not watch the road. It's right there in the name. Otherwise call it "Kinda Self Driving"
But given the accidents where people turn on FSD and go to sleep, clearly not everyone thinks that. And the better FSD is, the more crashes are avoided and the higher that percentage is.
First of all it informs you that it is not jet complete. There is a complete list of what features it has and what features it doesn't have.
You very clearly buying the right to have access to the newest features as the role out. This is very clearly spelled out when you buy it.
> But given the accidents where people turn on FSD and go to sleep
Do you have any actual data about people deliberately go to sleep?
I have seen no numbers what so ever to confirm what you are saying.
All I know is in my personal experience, everyone who bought FSD feels conned. Maybe you went into it eyes wide open, and if so good for you. Maybe they expected a faster rollout of features, maybe they didn't notice that part of the sales pitch at all.
How big is the disclaimer when you buy FSD? I'm not sure that any disclaimer is really worth the horribly misleading name.
>Do you have any actual data about people deliberately go to sleep?
Well, there was the one where someone went into the back seat to sleep. And there were several other high-profile crashes where people turned on FSD and passed out at the wheel at over 100MPH.
And, we only have data on when the FSD fails, and crashes ensue. Heck, we probably only have data on fatalities because the news doesn't want to get sued.
There are literally 100s of videos of the Autopilot doing an avoiding maneuver that prevented potentially accidents, some potentially deadly.
> Maybe they expected a faster rollout of features, maybe they didn't notice that part of the sales pitch at all.
I think its fair to say you expected the features faster, but the claim its not spelled out is false. If people don't read what is very clearly written on the website when you buy it, then that is their fault.
This is really to do with the expectations. Consider the safety belt, there is an expectation that with the safety belt on, the driver will not go through the windscreen in case of an accident. It is not the case that the driver has defeated the seat belt feature by not using the seat belt. A reasonable person will not have the expectation of having the safety provided by seatbelt without put it on.
Things are not as clear cut for assisted driving as can be seen in this thread.
This is false. Comma.ai/OpenPilot allows this. Video seems under-researched.