A 2025 Tesla Model 3 in Full-Self Driving mode drives off of a rural road, clips a tree, loses a tire, flips over, and comes to rest on its roof. Luckily, the driver is alive and well, able to post about it on social media.
I just don’t see how this technology could possibly be ready to power an autonomous taxi service by the end of next week.
Why was the driver not paying attention and why didn’t they just steer it back into the lane? It’s not called “Full Self Driving (Supervised)” for no reason. Hopefully Tesla get all the telemetry and share why it didn’t stay on the road, and also check if the driver was sleeping or distracted.
“Kill me” it said in a robotic voice that got slower, glitchier, and deeper as it drove off the road.
The car made a fatal decision faster than any human could possibly correct it. Tesla’s idea that drivers can “supervise” these systems is, at this point, nothing more than a legal loophole.
What I don’t get is how this false advertising for years hasn’t caused Tesla bankruptcy already?
What false advertising? It’s called “Full Self Driving (Supervised)”.
For many years the “supervised” was not included, AFAIK Tesla was forced to do that.
And in this case “supervised” isn’t even enough, because the car made an abrupt unexpected maneuver, instead of asking the driver to take over in time to react.
Because the US is an insane country where you can straight up just break the law and as long as you’re rich enough you don’t even get a slap on the wrist. If some small startup had done the same thing they’d have been shut down.
What I don’t get is why teslas aren’t banned all over the world for being so fundamentally unsafe.
What I don’t get is why teslas aren’t banned all over the world for being so fundamentally unsafe.
I’ve argued this point the past year, there are obvious safety problems with Tesla, even without considering FSD.
Like blinker on the steering wheel, manual door handles that are hard to find in emergencies, and distractions from common operations being behind menus on the screen, instead of having directly accessible buttons. With auto pilot they also tend to break for no reason, even on autobahn with clear road ahead! Which can also create dangerous situations.
Well, because 99% of the time, it’s fairly decent. That 1%'ll getchya tho.
To put your number into perspective, if it only failed 1 time in every hundred miles, it would kill you multiple times a week with the average commute distance.
Many Tesla owners are definitely dead many times, on the inside.
…It absolutely fails miserably fairly often and would likely crash that frequently without human intervention, though. Not to the extent here, where there isn’t even time for human intervention, but I frequently had to take over when I used to use it (post v13)
Someone who doesn’t understand math downvoted you. This is the right framework to understand autonomy, the failure rate needs to be astonishingly low for the product to have any non-negative value. So far, Tesla has not demonstrated non-negative value in a credible way.
What is the failure rate? Unless you know that you can’t make that claim.
Even with the distances I drive and I barely drive my car anywhere since covid, I’d probably only last about a month before the damn thing killed me.
Even ignoring fatalities and injuries, I would still have to deal with the fact that my car randomly wrecked itself, which has to be a financial headache.
That’s probably not the failure rate odds but a 1% failure rate is several thousand times higher than what NASA would consider an abort risk condition.
Let’s say that it’s only 0.01% risk, that’s still several thousand crashes per year. Even if we could guarantee that all of them would be non-fatal and would not involve any bystanders such as pedestrians the cost of replacing all of those vehicles every time they crashed plus fixing damage of things they crashed into, lamp posts, shop Windows etc would be so high as it would exceed any benefit to the technology.
It wouldn’t be as bad if this was prototype technology that was constantly improving, but Tesla has made it very clear they’re never going to add lidar scanners so is literally never going to get any better it’s always going to be this bad.
Saying it’s never going to get better is ridiculous and demonstrably wrong. It has improved in leaps and bounds over generations. It doesn’t need LiDAR.
The biggest thing you’re missing if that with FSD **the driver is still supposed to be paying attention at all times, ready to take over like a driving instructor does when a learner is doing something dangerous. Just because it’s in FSD Supervised mode it slant mean you should just sit back and watch it drive you off the road into a lake.
…is literally never going to get any better it’s always going to be this bad.
Hey now! That’s unfair. It is constantly changing. Software updates introduce new reversions all the time. So it will be this bad, or significantly worse, and you won’t know which until it tries to kill you in new and unexpected ways :j
To be fair, that grey tree trunk looked a lot like a road
It’s fine, nothing at all wrong with using just camera vision for autonomous driving. Nothing wrong at all. So a few cars run off roads or don’t stop for pedestrians or drive off a cliff. So freaking what, that’s the price for progress my friend!
I’d like to think this is unnecessary but just in case here’s a /s for y’all.
GPS data predicted the road would go straight as far as the horizon. Camera said the tree or shadow was an unexpected 90 degree bend in the road. So the only rational move was to turn 90 degrees, obviously! No notes, no whammies, flawless
It got the most recent update, and thought a tunnel was a wall.
… and a tree was a painting.
Took me a second to get it, but that’s brilliant.
I wonder if there might even be some truth to it?Wonder no more. Someone did this on YouTube using cardboard boxes, Tesla drove straight through them. Skip to around the 15 minute mark to watch it drive through the “wall” without even touching the brakes.
Edit: thought the person you were replying to said it thought a wall was a tunnel, not the other way round. Still funny to watch it breeze through a wall with a tunnel painted on it though.
Yes I know the video, what I was wondering is if it could be true that they tried to make the AI detect a wall with a road painted on it, and it falsely believed there was a wall, and made an evasive maneuver to avoid it.
I have visions of Elon sitting in his lair, stroking his cat, and using his laptop to cause this crash. /s
Why would you inflict that guy on a poor innocent kitty?
That tree cast shade on his brand.
It had to go.
HAL9000 had Oh Clementine!
Has Tesla been training their AI with the lumberjack song?
The worst part is that this problem has already been solved by using LIDAR. Vegas had fully self-driving cars that I saw perform flawlessly, because they were manufactured by a company that doesn’t skimp on tech and rip people off.
Lidar doesn’t completely solve the issue lol. Lidar can’t see line markings, speed signs, pedestrian crossings, etc. Cars equipped with lidar crash into things too.
I oversold it in my original comment, but it still performs better than using regular cameras like Tesla did. It performs better in weather and other scenarios than standard cameras. Elon is dumb though and doesn’t think LiDAR is needed for self-driving.
I wouldn’t really called it a solved problem when waymo with lidar is crashing into physical objects
NHTSA stated that the crashes “involved collisions with clearly visible objects that a competent driver would be expected to avoid.” The agency is continuing its investigation.
It’d probably be better to say that Lidar is the path to solving these problems, or a tool that can help solve it. But not solved.
Just because you see a car working perfectly, doesn’t mean it always is working perfectly.
Are those the ones that you can completely immobilize with a traffic cone?
A human also (hopefully anyway) wouldn’t drive if you put a cone over their head.
Like yeah, if you purposely block the car’s vision, it should refuse to drive.
The same is true when you put a cone in front of a human driver’s vision. I don’t understand why “haha I blocked the vision of a driver and they stopped driving” is a gotcha.
You say that like it’s a bad thing lol if it kept going, that cone would fly off and hit somebody.
Probably Zoox, but conceptually similar, LiDAR backed.
You can immobilize them by setting anything large on them. Your purse, a traffic cone, a person :)
Probably makes sense to be a little cautious with the gas pedal when there is an anything on top the vehicle.
That and if you just put your toddler on the roof of the car or something or trunk for a quick second to grab something from your pocket…VROooOMMM baby gone.
Yes lol
Why someone will be a passenger in self-driving vehicle? They know that they are a test subjects, part of a “Cartrial” (or whatever should be called)? Self-Driving is not reliable and not necessery. Too much money is invested in something that is “Low priority to have”. There are prefectly fast and saf self-driving solutions like High-speed Trains.
I have no idea, I guess they have a lot more confidence in self driving (ESPECIALLY Tesla) than I do.
Typical piss poor quality from Leon Hitler. F Tesla.
Don’t drive Tesla
I mean, if Elon was my dad, I’d probably have some suicidal tendencies too.
More like the abusive step-father
I use autopilot all the time on my boat. No way in hell I’d trust it in a car. They all occasionally get suicidal. Mine likes to lull you into a sense of false security, then take a sharp turn into a channel marker or cargo ship at the last second.
They have auto pilot on boats? I never even thought about that existing. Makes sense, just never heard of it until just now!
They’ve had it forever. Tie a rope to the wheel. Presto. Autopilot.
I’ll point this post out to Wall Street Bets, Maersk stock will pop 10%+ overnight.
That’s not how boats (ouside of hollywood) work, tho
They’ve technically had autopilots for over a century, the first one was the oil tanker J.A Moffett in 1920. Though the main purpose of it is to keep the vessel going dead straight as otherwise wind and currents turn it, so using modern car terms I think it would be more accurate to say they have lane assist? Commercial ones can often do waypoint navigation, following a set route on a map, but I don’t think that’s very common on personal vessels.
That’s similar to commercial airliners right?
Exactly. My car doesn’t have AP, but it does have a shed load of sensors and sometimes it just freaks out about stuff being too close to car for no discernible reason. Really freaks me out as I’m like what you see bro we just driving down the motorway.
For mine, it’s the radar seeing the retro-reflective stripes on utility poles being brighter than it expects.
Isn’t there a plane whose autopilot famously keeps trying to crash into the ground. The general advice is to just not let it do that, whenever it looks like it’s about to crash into the ground, pull up instead.
All the other answers here are wrong. It was the Boeing 737-Max.
They fit bigger, more fuel efficient engines on it that changed the flight characteristics, compared to previous 737s. And so rather than have pilots recertify on this as a new model (lots of flight hours, can’t switch back), they designed software to basically make the aircraft seem to behave like the old model.
And so a bug in the cheaper version of the software, combined with a faulty sensor, would cause the software to take over and try to override the pilots and dive downward instead of pulling up. Two crashes happened within 5 months, to aircraft that were pretty much brand new.
It was grounded for a while as Boeing fixed the software and hardware issues, and, more importantly, updated all the training and reference materials for pilots so that they were aware of this basically secret setting that could kill everyone.
The Being 787 Max did that when the sensor got faulty and there was no redundancy for the sensor’s because that was in an optional addon package
Even worse, the pilots and the airlines didn’t even know the sensor or associated software control existed and could do that.
Pretty sure that’s the Boeing 777 and they discovered that after a crash off Brazil.
“I’m confident that Save full self driving (SFSD) will be ready next year”
