In the piece — titled “Can You Fool a Self Driving Car?” — Rober found that a Tesla car on Autopilot was fooled by a Wile E. Coyote-style wall painted to look like the road ahead of it, with the electric vehicle plowing right through it instead of stopping.
The footage was damning enough, with slow-motion clips showing the car not only crashing through the styrofoam wall but also a mannequin of a child. The Tesla was also fooled by simulated rain and fog.
I hope some of you actually skimmed the article and got to the “disengaging” part.
Yeah but that’s milliseconds. Ergo, the crash was already going to happen.
In any case, the problem with Tesla autopilot is that it doesn’t have radar. It can’t see objects and there have been many instances where a Tesla crashed into a large visible object.
That’s what’s confusing me. Rober’s hypothesis is without lidar the Tesla couldn’t detect the wall. But to claim that autopilot shut itself off before impact means that the Tesla detected the wall and decided impact was imminent, which disproves his point.
If you watch the in car footage, autopilot is on for all of three seconds and by the time its on impact was already going to happen. That said, teslas should have lidar and probably do something other than disengage before hitting the wall but I suspect their cameras were good enough to detect the wall through lack of parallax or something like that.
That is like writing musk made an awkward, confused gesture during a time a few people might call questionable timing and place.
That’s so wrong holy shit
Don’t get me wrong, autopilot turning itself off right before a crash is sus and I wouldn’t put it past Tesla to do something like that (I mean come on, why don’t they use lidar) but maybe it’s so the car doesn’t try to power the wheels or something after impact which could potentially worsen the event.
On the other hand, they’re POS cars and the autopilot probably just shuts off cause of poor assembly, standards, and design resulting from cutting corners.
if it can actually sense a crash is imminent, why wouldn’t it be programmed to slam the brakes instead of just turning off?
Do they have a problem with false positives?
if it was european made, it would slam the brakes or swerve in order to at least try and save lives since governments attempt to regulate companies to not do evil shit. Since it american made it is designed to maximise profit for shareholders.
I don’t believe automatic swerving is a good idea, depending on what’s off to the side it has the potential to make a bad situation much worse.
I’m thinking like, kid runs into the street, car swerves and mows down a crowd on the sidewalk
Its the cars job to swerve into a less dangerous place.
Can’t do that? Oops, no self-driving for you.
I’ve been wondering this for years now. Do we need intelligence in crashes, or do we just need vehicles to stop? I think you’re right, it must have been slamming the brakes on at unexpected times, which is unnerving when driving I’m sure.
So they had an issue with the car slamming on the brakes at unexpected times, caused by misidentifying cracks in the road or glare or weird lighting or w/e. The solution was to make the cameras ignore anything they can’t recognize at high speeds. This resulted in Teslas plowing into the back of firetrucks.
As the article mentioned, other self-driving cars solved that with lidar, which elon himself is against because he says AI will just get so good and 2d cameras are cheaper.
This is from 6 years ago. I haven’t heard of the issue more recently
https://www.washingtonpost.com/technology/interactive/2023/tesla-autopilot-crash-analysis/
The tesla did not consistently detect that the thing infront of it was a truck, so it didn’t brake. Also, this describes a lot of similar cases.
I remember a youtuber doing similar tests, where they’d try to run over a fake pedestrian crossing or standing in the road at low speed, and then high speed. It would often stop at low speed, but very rarely stopped or swerved at high speed.
Wouldn’t it make more sense for autopilot to brake and try to stop the car instead of just turning off and letting the car roll? If it’s certain enough that there will be an accident, just applying the brakes until there’s user override would make much more sense…
False positives. Most like it detected something was off (parking sensor detected something for example) but doesn’t have high confidence it isn’t an erroneous sensor reading. You don’t want the car slamming on brakes at highway speed for no reason and causing a multi car pileup.
Normal cars do whatever is in their power to cease movement while facing upright. In a wreck, the safest state for a car is to cease moving.
I see your point, and it makes sense, but I would be very surprised if Tesla did this. I think the best option would be to turn off the features once an impact is detected. It shutting off before hand feels like a cheap ploy to avoid guilt
that’s exactly what it is.
Rober seems to think so, since he says in the video that it’s likely disengaging because the parking sensors detect that it’s parked because of the object in front, and it shuts off the cruise control.
It always is that way; fuck the consumer, its all about making a buck
I’ve heard that too, and I don’t doubt it, but watching Mark Rober’s video, it seems like he’s deathgripping the wheel pretty hard before the impact which seems more likely to be disengaging. Each time, you can see the wheel tug slightly to the left, but his deathgrip pulls it back to the right.