Oddly enough, when a Tesla accelerates at a barrier, someone dies: Autopilot report lands A Tesla with Autopilot engaged accelerated toward a barrier in the final seconds before a deadly crash, an official report into the crash has revealed. Apple engineer Walter Huang was driving his Model X P100D on a Silicon Valley freeway on the morning of March 23 when the car, under computer control, moved into the triangular …
Exactly.
Here's why it might have hit the divider:
https://youtu.be/6QCF8tVqM3I
And El Reg has (I think) already run a piece on why the car won't stop for stationary objects.
Here is WIRED's take on it:
https://www.wired.com/story/tesla-autopilot-why-crash-radar
And a summary feom elsewhere:
Tesla "Autopilot" cannot see stationary objects. It has a radar, but it is 1-D and low resolution. It uses this to localize other moving cars. In theory, it could use this to see a highway divider or firetruck it's about to hit, but since it's 1-D that would also mean that it would have to slam on the brakes when it sees an overpass coming up, because it can't tell the difference between an overpass and a highway divider or a firetruck. It can assume that overpasses aren't driving at 60mph, so it will see other cars. The Tesla algorithm is "look for lane markings and try to stay between them, and don't hit any moving vehicles. If there is any stationary object, including another a vehicle in your lane, the "Autopilot" will plow right into it.
I've said it before and I'll say it again. Its a glorified cruise control. Calling it "Autopilot" vastly overstates its abilities and lulls drivers into a false sense of security.
It does not really provide long term, hands free driving as Tesla say, yet it can manage long enough to remove any attention span you may need to do anything manually while you drop your coffee and put the phone down etc.
If they stopped calling it autopilot it will remove the inappropriate belief that the car becomes self driving.
encouraged by the name and marketing, owners think [...]
I've said it before and I'll say it again. Its a glorified cruise control. Calling it "Autopilot" vastly overstates its abilities and lulls drivers into a false sense of security.
It does not matter what people think, with an autopilot in an aircraft the pilot must stay in position, be ready to take over anytime. People SHOULD KNOW.
It is rather dumb that people think autopilot means autonomous driving when it does not. You are supposed to keep your hands on the wheel AND your eyes on the road. You can move your feet away from the pedals and enjoy the ride, HOWEVER, keep hands on wheel and eyes on the road.
Reminds me of the story of a camping car driver who thought cruise control was autonomous driving, engaged it and went to the kitchen for a nice hot coffee .... or wing mirrors which mention "objects in (the) mirror are closer than they appear" No f'ing shit Sherlock ...
You get a car, you work out what options it has AND what these do PRIOR TO USING THEM.
The world cannot save all cretins, we are all trying very hard, but, you know, some are just beyond help. You can rename the option to super cruise control or whatever, there will always be cretins who think it means autonomous driving.
"If they stopped calling it autopilot it will remove the inappropriate belief that the car becomes self driving."
It's a glorified adaptive cruise control with lane keeping assist. One of it's biggest problems is that it works well enough in many, if not most, average driving situations, but royally screws up from time to time. People get lulled into letting it do too much with too little attention up until the point where it hits the emergency vehicle.
"...I've said it before and I'll say it again. Its a glorified cruise control. Calling it "Autopilot" vastly overstates its abilities and lulls drivers into a false sense of security..."
I've hadn't really noticed that people had such a messed up definition of autopilot until the last few days when this crash's report came out.
This is Autopilot. Not a chauffeur, or a self-driving car. On and airplane, Autopilot holds a heading and an altitude. It does not steer around other planes or terrain. On a boat, autopilot will hold a magnetic/true-heading course, it will not avoid other boats or undersea obstructions and the shore. Sure other systems have been piped into those autopilots to do more advanced stuff - reaching waypoints can trigger the loading of the next waypoint with a new heading - and radars, GPS, TCAS, and transponders add to the overall information picture, but a person still needs to respond when all the alarms start going off.
If you set a heading on a boat and then go below deck to get a cup of coffee for 10 minutes, the autopilot will drive itself right into another boat and not think anything was wrong with that. We don't have <u>self driving</u> commercial boats and planes yet either... The US Navy had a few crashes in recent history that show what happens on a boat when you don't pay attention to what your automation systems are doing.
It seems that only in the delusions of people who are literally >>>dying<<< to get a self driving car, that the definition seems to have been misunderstood.
"No crossing solid lines. Should be pretty simple."
ISTR that the markings were mostly washed out and probably not easily recognized. The lane visually just expands to the left.
The "bug" was reproducible:
Simple for an experience human driver, apparently not for an artificial not-so-intelligence. The main theme in the article is correct: Tesla's use of the term "autopilot" was fatally misleading. Sad that a company that could have led the transition from fossil fuel to electric transportation may now be derailed by an irresponsible (juvenile?) marketing decision.
I believe Tesla get treated unfairly here. First, humans are also involved in stupid accidents and we accept that as fact of life. To expect a perfect record for self driving cars (or the 'autopilot' stages run by tesla, which can be considered precursors thereof) is not rasonable. And yes this new type of accidents will look stupid to us human overlords, but the human errors often look stupid too - read your news if you don't agree.
Second, this article makes a complicated case for why the autopilot is too blame (hands on or off the steering wheel,...). But traffic rules must be simple to follow, and the rule applying here is simple: the human driver must remain in control. KISS, otherwise you just feed the lawyers.
Finally a comment on battery fires. Extinguishing a battery fire with water does not work. Not a fault if Tesla. Looks like the firefighters have to get some training to prepare them for the electromobility era.
I agree with Kieran that Tesla blatantly oversells their cars and the autopilot function. Just be aware when you think about your next car purchase, but don't use it to build a criminal case against Tesla.
So Tesla has follow-distance control but no emergency stop? Isn't that becoming standard even on regular cars?
My Honda Jazz has it but only at low speed. I've had it trigger once and it was - sorta - right. Someone pulled out onto a roundabout in front of me. It would have been a cheeky but safe lunge if it wasn't for the fact they were towing a trailer. So I stopped (no panic, just slowed and waited). But I was a bit irritated and concerned about being rear-ended so when the trailer was half way across my front I accelerated. I knew the trailer would clear before I got there but the car disagreed. Cue lots of beeps, lights flashing on the dashboard and the brakes coming on.
But I think it only trggers below 30mph so wouldn't help in this scenario.
I'm 73 and receive a government pension, though I still have to work too. I read The Reg every day between sessions of designing machinery in Autodesk Fusion 360, building and programming projects for my Arduino boards, learning to play guitar, and riding too fast on my BMW K1200R. Though from your remark, I'm pretty sure that by pension age, you'll be too addled to comprehend The Reg, I'm not, and neither are a lot of other Reg readers. Stifle your ageist remarks. Sooner or later you'll be talking about yourself.
I'm 16 and receive pocket money, though I still have a paper round too. I read The Reg every day between sessions of designing machinery in Autodesk Fusion 360, building and programming projects for my Arduino boards, learning to play guitar, and riding too fast on my Gilera Runner 50. Though from your remark, I'm pretty sure that by pension age, you'll be too addled to comprehend The Reg, I'm not, and neither are a lot of other Reg readers. Stifle your ageist remarks. Sooner or later you'll be talking about yourself.
It does have emergency stop if (eg) the car in front of you slams it's brakes on, but as explained up thread, it might not be able to 'see' stationary objects.
That's ridiculous - my 10 year old Volvo can 'see' stationary objects and will warn about them loudly, as in windshield flashing red and lots of warning noises.... No automated braking as it's too old for that feature, but cruise control will dramatically slow down the car if engaged, including downshifting for engine braking.
Happens sometimes if you are in a long left turn lane cut out of a median and there is a control box or other square-ish object on the other end of the turn lane (but on the other side of the cutout) which you may be approaching rapidly as you reach the left turn point...
"but as explained up thread, it might not be able to 'see' stationary objects."
Tesla is explicitly clear that the autopilot is unlikely to detect and stop for stationary objects in its lane
when travellling in excess of 50mph
Autopilot is an enhanced cruise control. It's not a robot driver.
"So Tesla has follow-distance control but no emergency stop? Isn't that becoming standard even on regular cars?"
I hope not. That would be the first thing I'd want to rip out of a new car. Sometimes it's better to stay at the same speed or go faster and maneuver than to slam on the brakes. I don't see any sort of autonomous car being able to make that decision anytime soon.
Sometimes it's better to stay at the same speed or go faster and maneuver than to slam on the brakes. I don't see any sort of autonomous car being able to make that decision anytime soon.
Agreed. The automated panic braking ("collision avoidance", which really means "convert a collision at the front end of your car to one at the back end") in my wife's new Volvo is a huge pain in the ass. It triggers inappropriately all the time, generally when some idiot pulls into the lane in front of the car at too short a distance (in the computer's opinion). It's just luck that she hasn't been rear-ended yet by a tailgating vehicle when that happens.
Volvo's Pilot Assist features are highly rated by reviewers who like this sort of thing. I'd hate to try to drive a vehicle that has a low-rated implementation.
" If there is any stationary object, including another a vehicle in your lane, the "Autopilot" will plow right into it. "
If true, in my place, the day when everyone has this, deaths on road will rocket high vs. today.
You can find anything idle on the roads here: rocks, animals, idle cars of a random idiot having a phone call etc ...
" >> If there is any stationary object, including another a vehicle in your lane, the "Autopilot" will plow right into it.
If true, in my place, the day when everyone has this, deaths on road will rocket high vs. today."
How so? Cars today will plow right into anything you aim them towards. How would it be any different?
"If there is any stationary object, including another a vehicle in your lane, the "Autopilot" will plow right into it. "
Sorry, that last line confused me - how come it can spot a moving car/object in front of it and slow accordingly but not a stationary one? That makes no sense and suggests it would plow into a traffic jam which is clearly not the case since Autopilot works in traffic. I suspect the radar signal from the divider wasn't big enough at first to cause a reaction but you'd expect that when fairly close to it the computer would realise there's a stationary object in front and at least try to slow down. The fact that it didn't suggests a serious bug rather than an overall design fault.
how come it can spot a moving car/object in front of it and slow accordingly but not a stationary one?
The sensors can't see far enough ahead to spot objects off in the distance, and by the time those stationary objects finally come into range, at highway speeds, it's far too close for the vehicle to react and the collision is inevitable.
Essentially, Telsa's autopilot feature is like driving with Mr. Magoo*--or Carrie Underwood during a snowstorm.**
The future of self-driving / autonomous vehicles is to prevent those types of drivers from ever being on the road--but we're clearly not there yet, and Tesla isn't helping when it's designing and promoting systems as "autopilot" when they're far worse than most human drivers.
* - Mr. Magoo was extremely nearsighted / almost blind cartoon character that frequently got into hijinx as he would bump into things.
** - In her song "Jesus Take the Wheel" she hits a patch of ice and skids out of control. At that point she ignores any previous driving instruction she might have received, takes both hands off the wheel in order to pray for Jesus to change from his role as co-pilot to actual pilot, and steer her to safety.
Because tracking stationary objects is too hard for its pea-brained AI? I've driven on the Cali freeway system. It's not much different than NYC, but a lot easier to negotiate than, say, Philly. Anyone who uses autopilot (or cruise control, for that matter) in those kinds of road systems might as well point a loaded revolver at their head and hope the next chamber is empty.
Seeing the stationary objects is easy enough.
Determining which ones are actual hazards and which ones are overhead gantries, street furniture, potholes/cracks, leaves blowing across the road, stationary traffic in another lane or an actual threat is another matter entirely.
Lidar + processing power (as seen in autonomous vehicles) maps out the surroundings. With the 1D radar used for autopilot, there is no positional sense - at all.
You decide if you want to keep slamming your brakes on automatically and unnecessarily when on the freeway, causing someone to ram you up the rear, or filter out all the stationary returns.
Bear in mind that the driver is meant to be in control at all times, Autopilot or no.
Traffic slowing to a halt is easy, you can track the change in velocity and match it.
"how come it can spot a moving car/object in front of it and slow accordingly but not a stationary one?"
I wonder if there is a limit on detections (low-pass filter) to prevent falsing. It can see moving things going in the same direction because the differential speed is lower. When something is seen moving at 75mph, it gets filtered out as a false signal.
The autopilot probably accelerated because cruise control was set to a higher speed and the car in front made it slow down. When the car moved out of view, it sped up to reach its set speed. And it's not wrong to call it an autopilot. An airliner autopilot will happily fly straight into a mountain if you tell it to. It's just steering automation, not HAL-9000. But I agree that Tesla are jerks for blaming the victim, particularly when it should be extremely easy to detect splitting lanes in the map data.
"An airliner autopilot will happily fly straight into a mountain if you tell it to."
IMHO if you manually tell it to do a dangerous thing, it stops being an autopilot at that point. Aircraft autopilot follows routes, with set safe altitudes, and terrain-following radar to avoid collisions.
Tesla's tech shot off into a barrier.
C.
"I thought he worked for Apple, that may explain why he was driving it wrong."
I think you were downvoted for a variety of reasons, but I also think there's a sensible point to be made there. Apple have created an expectation of what electronic systems do, and I imagine it's shared by their engineers.
At any time, people's expectations are related to general state of the art. In a time when a lot of people had coal fires, it didn't seem odd that a steam locomotive needed someone to shovel coal into a firebox. In a world of central heating, it seems a bizarrely dangerous idea.
Whatever Apple's faults as a company (I'm not going there) Apple stuff does pretty much what it says on the box. If an Apple engineer read "Autopilot" as "pilots car automatically" it would be unsurprising. Transportation technology probably wasn't his thing, or perhaps he wouldn't have bought a Tesla. He wanted an Apple type experience, i.e. pay a whole lot of money for something and then expect to have it do what it seems to claim.
"Aircraft autopilot follows routes, with set safe altitudes, and terrain-following radar to avoid collisions."
Not true. Even advanced autopilot systems like those in a Cirrus SR type aircraft will happily fly itself into terrain (whilst screaming TERRAIN and flashing red at you if certain features are added/enabled) or even chase itself into a stall* if a pilot does not intervene. This is why these and most other aircraft have a big AP Disconnect button on the control yoke to disable the thing instantly.
I'm not sticking up for Tesla here, just defining what an "autopilot" actually is. Tesla need to sort that out. They really do.
*I've tested this "feature" personally. They insist on it on your first flight in the thing.
Being of an antipodean nature, my first thought here is "Mt Erebus" and "Flight 901" (not to mention "orchestrated litany of lies"). However, that was 1979 and presumably aircraft autopilots are now a bit more advanced.
Are there any more recent inadvertent controlled flights into terrain that might provide useful for this discussion?
Are there any more recent inadvertent controlled flights into terrain that might provide useful for this discussion?
Here you are. Looking at a few of the recent crashes classed under CFIT that involved modern airliners, they were all caused by crew ignoring or responding too slow or incorrectly to warnings.
"The Sukhoi scandal:"
The captain of the jet was Alexander Yablontsev (57), a former Russian test pilot;
Human factors to the fore again - and yet another example of why ex-military fliers are a poor choice for civil transportation. They tend to press-on regardless when anyone sensible and cautious would have diverted. Being able to safely land 95% of the time is one thing but cleanup after the last 5% is problematic and unlike a military aircraft the people sitting in the back didn't sign on for that risk.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
