Watchdog growls at Tesla for spilling death crash details: 'Autopilot on, hands off wheel' The US National Transportation Safety Board (NTSB) has expressed displeasure with electric carmaker Tesla for releasing information relevant to a fatal Model X crash in California last month without alerting the agency beforehand. The NTSB began investigating the killer smash last week. "The NTSB is unhappy because parties to …
Ok, so Tesla has a $3000 option on a $5000 enhancements . What extra safety features does that package use?
If none, why would I trust it?
If extra safety features are programmed into the “upgrade “ why are they not in the driver assistance pack?
Self driving cars have got to have safety first and foremost, the highest levels of protection baked into the base packages. Extra functions to aid driving richer people than me can pay extra for but safety must be paramount.
what "extra safety features" have to do with anything. The $3000 option is for the self-driving pack which, unsurprisingly, enables self-driving. It's software that does something useful and you pay for it. Are you arguing that self-driving is so much safer that Tesla are ethically obligated to provide the software for free?
Self driving cars have got to have safety first and foremost, the highest levels of protection baked into the base packages
Why? Most other safety technologies started off as extra cost premium features - safety glass, seatbelts, crumple zones, airbags, assisted braking, stability control, antilock brakes - all of those things first came to market on premium vehicles, and were only mandated to ensure that the trickle down of technology was extended to all vehicles.
Self driving cars will not be perfect, but if they already reduce casualties* compared to meatsacks, then delaying them further to insist on additional levels of safety will cost more lives than it saves, as a matter of simple maths.
* I'm not convinced by the studies that Tesla trumpet on the 40% reduction in accidents, because they don't look to be properly controlled comparator groups. Which doesn't mean Teslas are not safer, it just requires a more sceptical, rigorous and scientific study. My personal guess is that like for like a Tesla is safer than a peer group car and peer group driver, but only by an unimpressive single digit percentage.
Consider your family member is killed by a base model with “rudimentary “ safety features.
If your family member was killed by a car on autopilot. It had base level safety software and you knew more sophisticated software was in use in a higher spec of car. You would be looking to sue the pants off Tesla or whoever.
The grounds? Sub standard equipment would be a start.
Safety isn’t about the drivers it’s about the victims.
.... you knew more sophisticated software was in use in a higher spec of car. ....
Yeah, sure, why is an AI different from other things we can have Right Now? Like Better Drivers or well maintained vehicles?
I'd like to first see someone manage to successfully sue one of the transport companies for running lorries with unqualified / drunk / overworked / drugged drivers, which, unlike AI drivers, are causing carnage at least every month or so when they rear-end some traffic jam because they are sleeping or on their mobile. If one can pin some corporate responsibility for the flesh-bots in their service, then perhaps one could do this with the AI once this exist at some remote point in the future, perhaps. I would not hold my breath until this happens.
What extra safety features does that package use?
I think it simply bumps the value of a hidden "driver personal survival"-parameter, so that the MI in the autopilot is more likely to prioritise driver survival over that of the proverbial school bus full of nuns and children. Kinda like happened in "I Robot". Moar Dollar, biglyer Bump. But, I could be wrong.
As an engineer, if my car swerves more than twice at the same location on Autopilot, I'd have my fscking hands on the steering wheel while passing through that location.* I really wonder about his state of mind.
*- "Once is happenstance; twice is coincidence; thrice is enemy action."
Interesting about this ... when (I think) the Guardian did an article on driving a Tesla to a Gite in the middle of France (to see if it was really possiblte to use one for a holiday like this - answer was yes but with a bit of planning of the recharge stops needed) there article commented that the road markings around the exits on French autoroutes confused the autopilot feature and the car would often start to turn to exit the autoroute before manual intervention having to steer it back.
"the road markings around the exits on French autoroutes confused the autopilot feature and the car would often start to turn to exit the autoroute before manual intervention having to steer it back."
Thats why true autodrive will have to wait until true AI that can grok the whole world around it, not just markings on the road, comes about. At the moment these systems are little more than glorified line followers with (apparently poor) collision detect systems. Personally I don't see the point - if you have to keep your hands on the wheel most of the time anyway then how is it any assistance other than for the congenitally lazy? You might as well just turn the wheel yourself. In fact I'd find it more stressful keeping an eye on the automation AND the road ahead than just doing the latter in manual mode.
This post has been deleted by its author
That was a stupid argument when people first made it, and it is a stupid argument today. What a pilot thinks "autopilot" is is 100% irrelevant, because Telsa is not selling cars only to licensed pilots. What the average person thinks autopilot is is the only thing that's relevant.
It is kind of crazy that Tesla wants to keep that name that is becoming associated with "deathtrap" for the average person at this point.
I read many posts in another forum that were harping on how the autopilot is little more than technologies available from other companies but named differently like lane assist, auto braking etc. Someone posted a link to a google maps satellite view where the accident occurred, and the claim is the car somehow got confused thinking there was another lane and moved into that "lane" even though it wasn't a lane which then ran into a barrier.
What doesn't make sense to me is why the car would do that, at least I haven't read mentions of Telsa's (or other company's tech) behaving in that they pass cars by themselves (but maybe they do I don't closely track this stuff). Anyway if the car's basic function is to stay within the lane how could it possibly get confused of lines to the left of the car, it should see easily the line on the left side of the car is a solid white strip marking the boundary, and on the right side is a dashed line marking the boundary. Sure there is ANOTHER solid white line past the one closest to the car, and somehow the car thinks it should cross one solid white line to align itself with the next white line it sees. For me it all comes down to the logic involved in deciding to cross a solid white line on a highway, which at least the last 20 years or so of my driving on the west coast I don't recall ever there being a situation where you can "legally" do that (breakdowns excepted of course).
I've driven that stretch of road many times myself having lived in the bay area from 2011-2016(and travel back several times a year I don't live far away just far enough for lower cost housing).
I could probably understand if the weather was really bad, or debris on the road or something to mask the lines, but have seen no claims of anything like that.
The lane in question was quite straight as well, in the grand scheme of things(all the situations the car would face) it should be a simple situation for the car to stay within the lane on a mostly straight highway road with good road conditions and good clear weather.
I recall one time I think back in 2005 I was driving from Boston to Montreal on a Friday night it was in Feb or March, light snow.. I have very little experience driving in the snow. Anyway was in a rental car with my friend and we were in Vermont at the time, they salted the roads or something a lot, the roads were not slick(at all) but they were almost completely white. As in I cannot see the lines in the road. Not many cars on the road. My friend said stop driving like I'm drunk, and I wasn't drunk had no drinks that day. I was driving and trying to follow whatever lines I saw in the road, sometimes I saw the line on the right side of the car, other times I only saw the one on the left(so naturally went to both sides of the lane many many times as I tracked the lines). No accidents or near misses or anything but something that stuck in my head as to a time where I really could not see the lines on the road. I think after we got into Canada it was fine then, just a couple stretches of road in Vermont that were particularly scary (slowed down of course for those bits). One of those bits was directly before the border crossing. Fortunately the cops there had no issue with how I was driving.
Even more fun is driving on a flat expanse of white between two distant fences, knowing that somewhere under the snow is a road.
Late at night on a motorcycle, clear night with near full moon, and ground fog up to about half a meter high. On a local road with alternately ditches to the side(s), or fencing with barbed wire. Riding with only the parking light it was possible to make out the road when looking not more than a few meters ahead.
I don't think I did over 10 km/h that stretch.
Some roads in the Moors (Lancashire and Yorkshire) hove these tall wooden poles spaced down both sides of the rosd. Strange until you realise that in snow they are the only thing visible to show where the roads is.
A bit like the withys along tidal channels in rivers. They show you where the channel is at high tide when you can't see the mud inches below the surface.
https://electrek.co/2018/04/02/tesla-fatal-autopilot-crash-recreation/
"Now another Tesla owners tried to film his Model S following the same lane change scenario on Autopilot in an almost identical section of road in Chicago and it might show exactly what happened during the accident:"
"We can see the driver ignoring an alert to ‘hold the steering wheel’ sent out a few seconds before the barrier just like Tesla said in its report based on the logs – though that was likely a time-based alert.
Then it seems like Autopilot’s Autosteer stayed locked on the left line even though it became the right line of the ramp. The system most likely got confused because the line was more clearly marked than the actual left line of the lane.
That led the car directly into the barrier and it’s easy to see how a driver who is not paying attention couldn’t have been able to react in time since the driver who recreated it was barely able to apply the brake in time himself."
Which is all well and good, and very interesting in itself...
But what bothers me is why the autopilot didn't stop, or at least slow, the car when it perceived that it was rapidly approaching a solid obstacle, regardless of "lane markings". What if it had been completely right about the lane, but there had been a stationary car in it? Wouldn't it have stopped?
Or did it not perceive the obstacle? Because that's a whole other can of worms, but no less wriggly.
I'd imagine for the same reason that one went under the lorry - couldn't pick it out of the background. The concrete barrier was missing the crash thingybob on the front of it, so could have appeared as a concrete block against a concrete road.
Oddly my small sprog did something much the same the other day. He crashed his bike into a concrete wall because he couldn't pick it out against a concrete path.
"Oddly my small sprog did something much the same the other day. He crashed his bike into a concrete wall because he couldn't pick it out against a concrete path."
When a car in auto-pilot has about the same driving ability as a small sprog, I don't want it on the road anywhere near me, thanks.
Holy shit, that section of road is murder materialized! I saw the setting of the accident before so I knew what to look for and yet watching the "recreation" the first I realized anything at all is wrong was when the guy dropped the camera and hit the breaks. By the fourth re-watch I could see where the lane split marking were supposed to be, but they're dim as fuck, I would have probably completely missed them the first time around even driving personally. Yes, a fully self-driving vehicle would need to be able to detect that, but I'm not particularly surprised Tesla's glorified lane assist didn't. Crash barrier or no crash barrier, the absolute non-negotiable bare minimum that road needs there at all times is a long string of traffic cones in that "lane". And some actual fucking paint.
Sounds very similar to comment I made above taken from a Guardian article a year or so back which commetned that Tesla autopilot had often tried to turn onto exits rather than stay on the autoroute when they'd been driving in France. At least in that case it was just following the wrong bit of road but it seemed clear that it was recognizing the wroing lines as the lane marking.
"f the autopilot detects and warns for hands-off the wheel after 6 seconds, why doesn't it take further action if the situation isn't then rectified, e.g. by progressively reducing the speed by a safe rate?"
It does - but it first issues a couple of visual warnings,followed by an audible warning. After that, it slows, looks to pull off the road and stop.
The following videos show that Tesla's autopilot is deadly inadequate, merely a overhyped lane assistant:
https://www.youtube.com/watch?v=VVJSjeHDvfY
https://www.youtube.com/watch?v=KEaJW-Mk0Bo
https://www.youtube.com/watch?v=kDgJnlr6Ak8
Autonomous cars need a LIDAR. Tesla has none, their stereo-camera plus radar doesn't detect road conditions in a reliable way even in best weather conditions. How should it even excel at snowy weather in the Alps? Musk should be ordered a hefty fine, for doing very bad meaning PR work, constantly overstating the cars capabilities, intentionally naming it "autopilot" while being merely a half-blind lane assistant. Musk is teasing us, stop him. His greed not to add a LIDAR means a heavy road toll, real people die.
Until recently Tesla was using the Mobileye vision system for their Autopilot software. This system has several features, such as sign recognition, obstruction/vehicle/pedestrian recognition and lane tracking. The one at issue here is the lane tracking.
Lane tracking is a very treacherous capability. It's quite easy to develop a vision system that appears to work. It's easy with freshly painted lane markers and no exits or merges. Complexity and special cases quickly pile up with right hand exits and entrance merges. Quick fixes such as always following the left-side lane line come back to bite you when encountering left exits and lane merges. Roadways with faded and old ground-away lane markings that are confusing to human drivers are even worse for lane-following vision systems.
Tesla mitigated this by requiring each section of highway to be successfully observed several times before enabling Autopilot. I suspect that here we are seeing the weakness of that approach -- the roadway marking degraded below a threshold and the lane tracking made a bad decision. This has probably happened thousands of times before, with the harmless result of taking the wrong exit or driving in a break-down lane. Here it was a fatally bad result of driving into a barrier centered in the 'false lane'.
I think the whole idea on relying on anything that needs to be maintain by local authorities simply won't work for automated cars.
Because they are pretty much universally shite when it comes to maintain any roads that isn't outside their offices or just happens to be a road a councillor lives on!
In my neck of the woods there are streets without lane markings. They are painted on, wear off and are never repainted.
At the other end of the spectrum I know several roundabouts (am I right in thinking these are unknown in the US?) where there are myriad white lines, sometimes crossing live lanes, and single lanes splitting into two and all sorts of confusion for a human driver, especially one with possibly limited knowledge of the area.
And others where the designer appears to have settled on a bizarre layout just to see if he can get away with it.
Just the sort of things to make a (semi) autonomous car think "I want to go home".
At the other end of the spectrum I know several roundabouts (am I right in thinking these are unknown in the US?) where there are myriad white lines, sometimes crossing live lanes, and single lanes splitting into two and all sorts of confusion for a human driver, especially one with possibly limited knowledge of the area.
not to mention THAT 'magic' roundabout in Hemel Hempstead, single roundabout, surrounded by six other roundabouts, even the locals have issues ffs
Only those seeing it for the first time. Once you look at it, it's obvious it's a series of short carriageways joined by roundabouts; two lanes on the anticlockwise direction and one on the clockwise. For every combination of in and out, there's a correct lane to enter, and a correct lane to approach a roundabout to exit.
It's not difficult.
Mind you, where the A404 meets the A1... that's got clearly marked lanes but people seem to be completely incapable of staying within them.
In France, you have two types of roundabouts, some where any entering vehicle has right of way and others where entering vehicles need to give way. Some are a mixture of both where some entrances have right of way, others do not. The town where I took my driving test was full of roundabouts with priority to entering cars ...
Of course, right of way means there are no line markings, just, cars coming from the right have right of way to enter as they see fit ... how are automated cars going to cope with that ? Considering that since giving entering cars right of way is brain-dead, potential deadlock, most of these are now being converted to conventional roundabouts, but still ... how on earth is an automated car gonna get that right ?
Rush hour at the place de l'étoile in Paris, the automated car will wait until rush hour is over before entering the roundabout ...
icon: Paris, coz ... place de l'étoile ...
Roundabouts were unknown in my part of,the U.S. until a couple of years ago. Then the fascist city council dropped one on University Ave.at an inter change with Arlington Expressway. Much confusion and traffic jams ensued.The confusion continues to this day. I prefer the old 4 way stop junctions,
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
