Tesla FSD faces yet another probe after fatal low-visibility crash Tesla is facing yet another government investigation into the safety of its full self driving (FSD) software after a series of accidents in low-visibility conditions. In its latest opening resume [PDF] into a Tesla FSD investigation, the National Highway Traffic Safety Administration (NHTSA) said that it's taking a look at …
"If a few are good, more must be better, and too many should be just about enough..."
All of the inputs need to get through a Kalman filter and be assessed/weighted and then processed and a decision made. Too many inputs might bog the system down so there will be an optimum number and type. That may change with conditions as well as how the data is processed. If a microphone hears a siren, perhaps it's not a bad idea to process visual camera inputs to detect lights from an emergency vehicle. When we get a puncture, we sense the car biased towards the flat tyre, the noise and the change in performance. Just a reading from a pressure sensor (that can often go wrong) won't do it. It's the totality of disparate data that leads us towards the correct assessment.
Well yes, Musk's ego doesn't help. But there's also the issue of cost. Musk seems to believe, very likely correctly, that in order to dominate the vehicle market, he can't depend on just his current customer base -- the rather small segment that has more money than common sense. He needs to compete on cost. Lidar costs money. As does the software to process the results. And integrating the Lidar with the visual sensors will cost a bit as well. What do you do when the two sensors tell you conflicting things? Musk wants his vehicles to be inexpensive. More sensors don't make the vehicle cheaper. If anything ...
There's also something that people, especially bright people I think, are subject to. OMFS -- One More Fix Syndrome. People, for the most part, get by just fine with just their eyes. And they really don't need binocular vision for distance estimation. Because human eyes are so close together, that only works out to 5 or 6 meters. Teslae have "eyes" and "brains". No damn reason they can't work as well as people. Or better. Maybe much better.
OMFS is incredibly seductive. I've certainly encountered it many times in the past six decades. I've bought into it myself at times. And I should know better.
Just one more fix or maybe two -- can't be much more than that -- and FSD will be as good as Tesla has been claiming. LIke I said ... Seductive.
"Teslae have "eyes" and "brains". No damn reason they can't work as well as people. Or better. Maybe much better."
Eventually, yes. Probably. Possibly. :-)
Currently, not so much. We really don't know how a human brain works, let alone how it uses what it sees, merges that with past experience and "intuition" and whatever else the brain does *in an instant*, that computers attached to cameras simply cannot and possibly never will be able to do. Even with Lidar, I wonder how many self-driving cars could cope well with a busy roundabout? If Teslas already have issue with intersections, what hope have they at a roundabout? Waymos seem to get confused by roadworks and road cones.
Hell, man. *I* can't cope with a busy roundabout. Especially an unfamiliar one. Like all sensible people, I HATE the damn things.
A Cybertruck, like the proverbial 900 pound gorilla, can probably just proceed. Really, who or what is going to argue with it? Maybe the software in more vulnerable autonomous vehicles will just find routes that avoid roundabouts.
"Hell, man. *I* can't cope with a busy roundabout."
Good! It's people like you that let me get across easily, quickly and safely because I can tell those who hesitate and make a quick decision to see if I can get through the gap in traffic :-)
On a more serious note, yes, it really does take a human brain in those sort of situations to minimise disruption and slow-downs at roundabouts. Small, sometimes almost imperceptible signals such as which way a stopper driver is looking, do they seem to be hesitating, have they got Learner plates (or P plate) on, are they only looking at the roundabout traffic and not the traffic joining at the previous entrance to their right (left in some countries) etc. I've only ever seen one video of a self-driving car on a roundabout and it was painful to watch! Others have, in the past, talked about negotiating who backs up on very narrow country lanes. Part of my route home includes a residential street with cars parked on both side leaving not quite enough room to pass oncoming traffic if one or both vehicles are wider than a standard car. It's also a bus route. Can I get through and into the next spot where there's no marked cars before the bus gets here? It's a game with no fixed rules because the rules are different for some people :-)
As an ex-learner motorbike driver many years ago I once had the situation where I was waiting for a gap in the traffic to get onto the Holland Park roundabout (the one with the water tower) when I found myself moving forward. A black cab driver behind me had seen a gap, but wasn't paying attention properly and started to propel me forward. Luckily no damage done, but would expect better from a black cab.
Oh, I loved that one when I worked in Swindon! Very educational, but also remarkably easy to navigate once you understood it's not one roundabout so you treat each as its own entity :)
It's probably changed since I worked down there (early 90's) but it was a very good introduction to handling multiple linked roundabouts.
Couldn't be bothered to look up the cost of automotive lidar, eh? Try it. Google will be happy to help you out. As will other search engines I should think. If you think it'll be cheap, be prepared for a surprise. Or ... Here's an article. https://cdn.neuvition.com/media/blog/lidar-price.html (Is it going to be necessary for autonomous vehicles even if it boosts the vehicle cost by 3-4%? How the hell would I know?)
Unless my reading skills have abandoned me, that article says that now they've figured out how to do solid-state LIDAR instead of expensive spinning mirror things, it costs about $1000 to add automotive LIDAR. That is about the same cost as upgrading your Model 3 Tesla from White to a fancier colour like er, black. So would hardly seem to be a dealbreaker?
I believe this is where we need to insert the obligatory reference to ACC’s “Superiority” - https://www.baen.com/Chapters/1439133476/1439133476___5.htm
Despite being a a Clarke fan since childhood (after borrowing a nearly 20 year old copy of “The Wind from the Sun”, a fabulous collection of short stories, including the still worrying “Dial F for Frankenstein”) I only came across this a few years ago, but I feel it should be a required part of the curriculum for many, many professions…
Kind of says it all about how much anyone should trust Tesla's FSD, direct from Musk himself.
https://www.independent.co.uk/news/world/americas/us-politics/elon-musk-dominion-marjorie-taylor-greene-b2631928.html
>If a few are good, more must be better, and too many should be just about enough...
Not necessarily.
The trouble comes in the weightings one gives to each sensor; something like a LIDAR might on the face of it be really reliable, a RADAR less so. The trouble is that none of the sensors are totally reliable, and a "good" one that gets fooled outweighs the "less good" ones that are screaming "stop!!!".
There's a lot of similarities with biometrics; a mix of sensors of varying performance, and a conflict of requirements. A biometrics system wants to be good at recognising the right person, and good at rejecting the wrong person. With non-AI weighted combinations of sensor outputs, it's mathematically possible to optimise such systems for both requirements. And, as all AI systems are doing is adding things up with weights, they're also somewhat conflicted. And, if it comes to that, humans suffer the same problem.
And with FSD, there is also conflicting requirements. The car's got to go when it can, but must stop when it should. You can't make it too keen to go elsewise it'll be ignoring red lights, and you can't make it too hesitant.
Tesla's FSD
The proof as ever is in the results. And I think it fair to say that Tesla's results are sub-par, and no where near being true FSD. Adding a LIDAR would probably help, but as Waymo in Phoenix know, that's not going to magically make everything work properly either.
It's worse than that. The human eye works far better than the camera in most of those situations.
Eyes have an absolutely massive dynamic range, a much better peak angular resolution, and a ridiculously wide field of view to detect movement and siccade to identify it. Plus stereo depth perception, though it's less important at driving distances than you might think.
You can also be sure these cars have the lowest cost cameras possible.
"and a ridiculously wide field of view to detect movement"
I wish I had a link to point to, but while we do have a wide field of view that can detect movement, we have a very narrow perception of details. Our brains make up stuff constantly to fill in. With lots of experience, those invented stories are accurate enough to be getting on with but computers can't do that. It's our brains that we "see" with and even somebody with sight in one eye can sort out depth well enough to drive. What can be important is seeing that traffic is coming to a stop up ahead and it would be good to start slowing now rather than slamming on the brakes just before we pile into the tail end of the tail back. We can also tell if it's a lorry across the road up ahead or we are seeing an overpass at a certain point of view that's obscuring the roadway below.
Yes there are issues with the way human brain processes images - generally that makes us miss things, see What an RAF pilot can teach us about being safe on the road to understand the issue. But computers aren't ready to replace the human divers in most situations.
"A lot of drivers slam on brakes at the last second as they are taught that, and not to look up the road and anticipate"
Hmm, I was taught the opposite but I've ridden with others that are either accelerating or hitting the brakes. The two main problems with that is poor gas mileage and lots of brake wear. I would guess that they weren't taught that formally, but didn't get any instruction not to do that either.
I find it handy to see what the motorway is doing ahead and if I should turn off and take other roads when traffic is bad. Even if there is no good way to get around a closure, it's much better to be waiting in a cafe than being parked on the motorway with everybody else if there is no way to get turned around past that last exit.
This!! We were taught to remain situationally aware, by not only looking ahead, but by keeping your head on a swivel, looking at *all* your mirrors, your blind spots, and not getting distracted.
Another thing we were taught was to (if possible) look through/past the cars ahead of you to stay aware. That stops you from having to slam on the brakes or do some utterly uncontrolled manoeuvres to avoid getting squished.
Im not actually sure they are you know.
My ex passed her test quite some time after me and although I have done emergency services driving, even discounting this, theres so much she simply wasn't taught learning that I was. Silly stuff like reading the road as others have mentioned. If it was just her (and she is an awful driver) Id put it down to her but it really does seem to be a thing.
My instructor always said you are only being taught to pass the test. Once youve done that you learn to drive. I wonder if this is even more the case now.
Theres also a huge gap in skills between those that enjoy driving and those that HAVE to drive.
In my innocence I had assumed that _everybody_ drove like that... but then I grew up and it's clear to see that an awful lot of drivers - I hesitate to say the majority, but some days it feels like it - take absolutely no notice of anything other than the vehicle immediately in front of them (or possibly their phone).
An often-observed style in Germany is people on the derestricted autobahns driving behind people in the same lane and jamming on the anchors at the last moment, rather than lifting off and slowing down before they get to the slower vehicle in front of them. I was taught that brake lights on a motorway is always an indication of an emergency... yet here, my local car servicing place is amazed at how often I don't need new brake discs and pads.
The military term is situational awareness, and it's something that (a) takes time to build up and (b) must be kept up to date to have any value.
It's the (a) bit that puts the lie to the theory that FSD can hand back to a driver and all will remain safe - if said driver has been believing the FSD hype, he or she will not have any up to date situational awareness when the car hands back, and as that is classically when things are about to go wrong it merely provides a pass to Tesla to blame the consequences on the driver. I would love to see someone getting access to the stats at Tesla and discover there's a small time gap between handing back and an accident.
As for driving, I've had various courses, and especially defensive driving teaches you the value of anticipation - and planning for when you got that wrong (it happens, not everyone has a consistent driving pattern). The result is much smoother driving, and much safer.
Driving instructor here… the “anticipation” is the “P” in “SIPDE,” which stands for scan - identify - predict - decide - execute.
Which is asking a lot from a robot driver. Sure, it can scan (if a bird hadn’t soiled the sensor) and it can identify (it might distinguish among a few known shapes).
But as Python had taught, nobody expects the Spanish Inquisition!
(I feel like Tesla doesn’t have a clear idea behind its FSD technology: what is their vision, is it HAL-9000? Is it Christine? Maximum Overdrive? “The Car”??)
Tesla's FSD has the "P" too, but there it stands for "Panic" - the point where it quickly hands back to the driver.
All these drive systems from "let's force you in the middle of the lane even if it makes you break the law" lane 'assist' to FSD tend to be purely operational, barely tactical and there's certainly no strategy behind it.
I have adaptive cruise control in my car, which basically translates as "find any excuse to slow down, and do so by using the brakes". Whereas I would play with the margin to the next car to let the car coast (sort of, it's more taking the foot a bit off the gas/electric eel), no, this thing spots something in the next town and immediately goes for the brakes. Nice wake up jolt for the driver behind me (on a motorway, braking is Not Good News) whereas it is wholly unnecessary.
It's just not very impressive from where I stand. Rubbish user interfaces, touch panels where you should have physical buttons and the constant need to take your eyes of the road (where they ought to be when you're driving) to wade through some menu to kill off all the nuisances the manufacturers deem essential to bother you with.
It makes me miss my first car. It was simple and drank fuel like a bored priest, but it always started and it did exactly what I told it to do, even if that was a Very Bad Idea (if you have a 1.4l engine, muddy meadows are not for you). But it was fun, something I miss in that mandatory automated thing the company gave me.
"yet here, my local car servicing place is amazed at how often I don't need new brake discs and pads."
Hah! Same here! I've been know to get through one or even two further services before the "keep an eye on your brakes" message becomes, "yeah, they're getting a bit thin, time to change them" :-)
> Another thing we were taught was to (if possible) look through/past the cars ahead of you to stay aware.
I was taught that, and it is a very useful technique -- now rendered almost impossible by the huge SUV things that are bigger than a 1980's Transit van but apparently necessary to deliver a small child to school. And just to make doubly sure, the designers are now fitting them with extremely bright wall-to-wall. "lightbar" brake lights that give you no chance of being able to see past the car in the dark.
...and windows coming, from the factory, tinted so dark you can't see through them. Between that and the monster Fx50 pickup towering over me, it's a wonder I can see anything. Pillars are getting wider, as well...to hold those side airbags. And since we all have backup cameras now, no need for any visibility to the rear, right?
I do notice one positive. (some) car manufacturers have realised that placing the turn indicator right next to the headlight, while making for a lower cost lighting assembly, makes the turn indicator invisible at night (or when driving in daylight with the lights on). Some cars are now coming with combination running/indicator LED lights, which run white most of the time, but change to amber and blink as a turn indicator. Others are placing the turn indicators further away from the headlights, or killing the LED headlight when the indicator on that side is activated. Same thing is happening in the rear -- I have noticed several cars with low mounted turn indicators, away from the brighter brake lights, which tend to overpower them.
There have been some terribly poor human factors decisions by car manufacturers...if they considered them at all (e.g.: displays replacing buttons)
In the UK at least - and I assume the EU - there are very detailed restrictions on where lights may be placed, and when they must be used. https://www.legislation.gov.uk/uksi/1989/1796/contents
Though apparently most people don't bother using any of the light switches, and rely on the automation to do it for them (leading to the ridiculous situation of cars unlit in bright heavy fog, because there's lots of light, right?). And as for indicator switches? Don't get me started...
As a driver of said SUV type vehicle…believe me, I’d rather not be driving one, or paying the road tax and “luxury” tax on top of that, or be eating the large fuel bill that goes with it. However, due to a number of factors including the fact that the bodywork of the car has got thicker to take account of all the extra crash protection and airbags so there is less space inside, and the fact that I _have_ to have car seats for all three of my kids, it means I’m limited by the size of the seats to around seven vehicles I can buy - and trust me, I’ve tried pretty much everything out there apart from a Ford Galaxy, because $fordDealers…
many drivers are taught to touch the brakes on every curve even if they can see the road and verge ahead clearly. I give those drivers extra space in front of me (until some jerk decides to fill it with their guzzler)
Those drivers will struggle when they get behind the wheel of an EV. The regenerative braking that is standard on them means that you don't have to use the brakes in most situation.
I agree about waiting in a cafe than being stuck in a queue. As long as that cafe isn't at a Motorway services. Forgetting to pay to park will give you a real headache in the shape of a big fine.
You won't like adaptive cruise control then.
If you think the driver in front of you is a demented moron who seems to be using his brakes in such a manner that the brake lights look more like a disco light show, think again.
There is a probably an adaptive cruise control in play.
Some do, I think. Don't have that many new cars at my disposal. Last three were all Toyota: Camry, Swace (both with annoying ACC) and a GR Yaris. Latter hat ACC, but I did not get around to use it :)
My own cars come with regular CC (Scooby BRZ) or without (07' Impreza STI).
One of the reasons the traffic police in the UK had to stop using Volvos - you can't turn off the safety systems which prevent the cars from crashing into things (hence no ramming vehicles where required for public safety) and Volvo refused to provide a police modification with the software off.
BMWs on the other hand were great for the police until their engines started spontaneously combusting. Now the police seem to favour Skodas for traffic cars in some areas.
Lane assist is worse than adaptive cruise control... I can appreciate it coming in handy in the dark or when you're tired, but when you're tired, you shouldn't be behind the wheel!
The sudden interventions by lane assist are infuriating, especially when it *thinks* you're crossing into another lane, or when you are actively trying to move lanes and it gets it wrong and tries to resist. I also turn LA off as soon as I start the EV I regularly have use of. It's "no, piss off, I don't want you".
ACC in an EV is a lot more useful, but if regenerative braking requires ACC to actually engage the brakes, then maybe I need to turn that off too.
I fear that in the UK, at least, there was a change between being taught to drive by anticipating what’s happening, and being taught to drive by reacting to what’s happened.
When I did my PCV licence, I’d been taught to drive by my Dad, who’d been a coach driver for about 20 years at that point, who would observe what was happening all around his bus, and do things like work up and down the gears to suit the situation. When I then went to do some lessons with an instructor to polish up before my test, all that went out of the window - it was a case of reacting rather than anticipating, and one gear change per hazard - so instead of dropping down as the bus slowed off a motorway junction for example, you would stay in top gear until you could see if you needed to stop (top to second) or could carry on (top to third, perhaps).
The hell of it was the instructor wasn’t that keen on that driving style compared to what had gone before, but that’s what you needs to do to pass your test. It’s entirely possible now that there are people learning to drive that were never exposed to that level of anticipation based driving…
Using engine braking isnt taught anymore. We were explicityl told in 95 TO use engine braking as dumping the cluth meant you were no longer in control.
As an EMS driver and recently having been a spinal emergency in the last few weeks you apreciate enghine braking can give a much more controlled stop.
The argument is, I beleive, that engine braking shows no lights to the rear and increases more expensive to correct clutch wear.
And another factor. If an experienced human driver finds him/her-self looking straight into the sun, they will probably involve some strategy to deal with the situation. The simplest is simply to move their sensors (eyes) right, left, up, down, whatever. They may use a fold down sun blocker that most vehicles have mounted above their front window. Or they can invoke a filter (e.g. sunglasses). Or if it's safe, they may simply guide the vehicle a bit to a spot where the glare is less of a problem. Cars probably can't do easily do any of those things other than moving the vehicle.
Another possible issue, that really didn't occur to me until now. Vehicle cameras need to have a wide field of view. Vertically as well as horizontally if they are to deal properly with overhead traffic lights. (And there really are intersections near where I live that have only an overhead light in the middle of the intersection controlling traffic flow.). You aren't supposed to aim digital cameras into the sun because if the sun is anywhere in the camera field of view, the camera lens will focus its rays to a VERY hot spot that may damage some of the pixels in the camera's light sensor. Cars don't have the option of never facing into the sun. Are the optical sensors in vehicles going to turn out to be wear items that need replacement (and recalibration?) every X thousand miles?
Given that Musk is friends with Trump, he deems looking straight into the sun probably perfectly normal and so will most likely ignore the risk to the cameras.
Bad jokes aside (yes, won't give up the day job), you're right. If these cameras have to spot overhead traffic lights they would indeed get properly cooked by sunlight. Interesting point, and a good question.
How would the vehicle's brains note it has literally acquired a blind spot and how would it handle that?
We can also move our head around to get the best use of the sun visor just as I had to do over the weekend when the sun was very close to horizontal in front of me (rare event here in rainy UK!)
It really did get me thinking how a FSD car could possibly cope with these conditions, as for roads where the road markings have been just about erased by traffic how do the "lane keeping" systems of today cope? Thankfully I still have a 2012 dumb car apart from the cruise control.
If the Elongated Muskrat won't end the idiotic camera-only failure, NHTSA needs to step in to recall and suspend the sale of Turdlas with any feature that allows the car to have any level of control beyond full human driving.
The camera-only approach is a failure, and should not be allowed on the road.
I like Mark Kermode's name for him: Emo Musk(rat)
Musk is brilliant. Brilliant front man and a complete tool. Almost all these 'big' names are front men. We all know about Zuck and the FBI.
I was similar to Musk in a roll of successess that gave me complete control over my career. And it works a wonder unitl you get to the stage that
Musk got to a few years back with Twitter. When you have complete control and no boss and are very creative, as he is in his way which is a value, it rarely ends well.
The only saving grace and the thin slice of empathy I have for the man is that I ask myself how well I would perform in his role(s). Tesla was luck with Machiavellian politics. SPaceX was his real zentih.
“The camera-only approach is a failure”. Agree.
“If the Elongated Muskrat won't end the idiotic camera-only failure, NHTSA needs to step in to recall…..”
Disagree. NHTSA should step in to recall, if and only if the safety outcomes are inadequate. It’s really important to regulate outcomes, not technical solutions. The problem with that, it assumes that “we know the right technology answer”. I don’t, you don’t, and neither does the NHTSA. Maybe LiDAR, maybe radar, maybe something else entirely. The *worst* thing is to agree a tickbox “as long as you’ve done X, you’re all good”.
Personally, I’ve long been an advocate of “external aids”. Having a drop-in replacement for the human-piloted car, is certainly cool, with all the AGI implications; but is it the best way forward, and is it actually necessary to solve all the hard problems? Making visual recognition of road signs bulletproof is a hard and expensive algorithms problem…..but instrumenting them with a Bluetooth transponder can be done for under a dollar per road-sign. A car is not a faster horse, and the autonomous vehicle doesn’t need to do everything a human does, to be the answer we need.
A phased rollout of autonomous vehicles is entirely credible. First, external motorway radar/LiDAR infrastructure every 100 meters which gets us autonomy-on-the-motorway could be as cheap as $5M/km, paid by the manufacturer….It’s peanuts compared to the cost of the road, and we’re quite happy to do it with railways. Next could be city centres, the exact same place where we seem to be happy to spend to put numberplate recognition cameras, to enforce congestion charging. Next, all urban areas - by which time, you’ve served 90% of the worlds population; and by a happy coincidence is exactly complementary to motorways+railway. Address 80% of the need, by solving 20% of the hard problems of “massive variety of real-world environment”.
You may be underestimating the costs and are certainly underestimating the likelihood of co-operation between manufacturers, let alone governments. Car manufactures are already heading down the road of subscriptions, Car as a Service, uniquely paired parts replacements etc. :-)
Costs: the U.K. has a road network of 420,000km, 41million vehicles. That’s a rather surprising one vehicle per 10 meters. Or to put it another way, as soon as the sensor separation required exceeds 10meters, it’s cheaper overall to fix them on the road and broadcast the info, than mount them in the car. It’s not as crazy as you first imagine. That ratio is much less country-dependent than you might think: even the vast stretches of the USA, it works out to one vehicle per 23 meters; Germany is 18 meters, India is 19 meters.
Co-operation is a problem, or more precisely getting the car companies to pay the government a tax for cost-of-sensor avoided. Once road sensors are installed and the info is broadcast, on an open standard, then car companies would be foolish not to implement a $10 WiFi receiver, plus sensor fusion, to avoid the cost of a $1000 LiDAR.
This post has been deleted by its author
How many of those vehicles are on the road at any one time? Did you factor that into your calculations? What about how much each individual road is used? Surely there's no point putting expensive sensors into a country lane that is 1 car wide which only gets 2 vehicles an hour down it.
> but instrumenting them with a Bluetooth transponder can be done for under a dollar per road-sign
The hardware may well cost less than a dollar per sign, but someone has to configure each BT unit with the location and "direction" of each sign so that, for example, a speed limit sign at the entrance to a road is only acted on by vehicles turning into that road, not just driving past the end of it.
So each sign has to be surveyed by a highways engineer before a lower paid, lower skilled person can stick the Bluetooth unit onto the sign. And then, finally, someone else has to check that the right sensor got put onto the right sign and that it works as required.
If the solution involves doing something to every sign in the country, then a better approach might be to add the locations and details of all signs into a map gazetteer so that the vehicle can decide for itself whether it saw the expected sign or not.
Or maybe even just have a discreet QR code identifying the sign.
Probably a better solution for signs specifically, I agree. I’m agnostic about the actual optimum solution.
I’m pushing back against the idea that unless the car can do *everything* that the human can do, in an un-cooperative environment, then it is somehow “stupid” and inadequate solution.
Then you also have the problem of maintenance. You need to know when they've gone bad, been damaged via accident etc. or you'll have blind spots in this fancy network.
Now take a look at the state of the roadways within 10 miles of where you live, and think about the ones that are in the worse shape. That's the kind of shape parts of that fancy road network you want to build will be in. No thanks!
Maintenance is just as much a problem for in-car based solutions. We’ve all seen wrecks being driven around, and the MOT is only carried out once per year; and has little real value.
I agree it’s an issue. I’m not seeing why it’s *more* of an issue if the sensors are roadside, rather than needing the owner to service them regularly.
Are you concerned about the ownership of the risk? I suspect that private cars end synchronously with human-driving. For all practical purposes, they’ll all be taxis anyway. It’s not ideal, but it’s my best guess. Parking, charging, shared cost of ownership etc.
No, cheap immigrant labour will probably clean the taxis.
The dystopian nightmare we already live in, is that drive-through car-cleaning robots have existed for sixty years, but mostly disappeared when it turned out that immigrant hand labour was cheaper.
I never claimed the future would be better than the present.
Car cleaning robots have never existed. Automated car washes clean the OUTSIDE of the car, not the inside which is what passengers care about. Why should I care if the outside of a car is covered in dirt and grime if I'm just a passenger? So long the cameras etc. are clear so the autonomous driving software can work, it could be caked in an inch of mud for all I care!
I think you’ve missed my point: hand-cleaning the outside costs X; and that’s what a car-wash automates. Hand-cleaning the inside, the humans charge roughly the same price.
If the worlds simplest robot fails to undercut a human on the task of washing the outside, a complex robot can’t undercut for the work inside. The task will be done by cheap labour.
Now, if you want to *change the system we live in*, to guarantee humans a better wage at the low end, I would agree. But until then….
Poorly sighted people wear glasses. And if they're not confident in what they can see, they slow down.
Once upon a time, only a few months after passing my driving test, I found myself obliged to drive on unlit country roads through thick fog at night. The fog was so thick that, most of the time, I literally couldn't see the road at all. But I knew the roads very well. And it was an emergency - a relative was dying.
So I went on at a walking pace, feeling my way by the tilt of the car, until I reached my destination and, as I'd expected, the fog in the town was much lighter. It was an adventure, but I consider I minimised the risk by driving so slowly. If a Tesla managed to kill someone, it must have been going way too fast for the conditions.
And it's a fight to get them stop driving as they approach being legally blind. So people kill themselves and each other in large numbers under similar conditions. Hence the large number of "turn your headlights on during the daytime" sections over here.
The core issue here is, as it has been over and over with Telsa in recent years, in them implementation more than the approach. The car should have warned the driver and kicked out of self drive, if necessary slowing down till they took over. Better camera's may increase the window where the the system can see slightly, but the failure here was in the logic of the system's operation. It failed to judge the cutoff, or anticipate things like fog by using additional information like a weather report or info from other cars, or that it was about to point itself into the sun. I have an app on my phone that does that calculation and plots it into an AR view in real time. Musk is rushing to save his tanking stock price and cutting corners again.
Lidar wouldn't ever let a car safely operate consistently under those conditions, or operate on it's own without other sensors. It might let it operate slightly less unsafely for the brief period it's screaming at the driver to take over or to stop. In either case the software of the system needs to be able to reliably tell when it can't operate safely. Tesla is cutting corners again, which shouldn't be tolerated, but that is a problem with their implementation more than their approach.
Waymo cabs with way more sensors aren't magically immune to these problems either. If anything, they are facing more cases where if one of their sensors see's an anomaly the vehicle stops dead or does something stupid. That said Musk is nuts if he thinks he can make a camera only driverless car work in San Francisco year round.
Lidar, radar and cameras all have different trade offs, but if a heavy fog rolls in, or in smoke, it may only be able to safely operate for a few more seconds. Other techniques can do alot more in these conditions, that have little to do with the existing on vehicle sensors. The car should be operating off an accurate 3d model of the road. It should be getting a feed on oncoming vehicle locations from beyond line of sight. It should be getting accurate information on road and safety conditions in real time.
Instead we face-planted on vehicle to vehicle communications. The car is trying to use a ML model to "read" the road on the fly in most cases, even though the companies are buying high-res road data to use in their model training.
In a saner world, these cars would only be operating on a dedicated right of way until we sort those things out. Instead we are allowing some genuinely terrible managers to push dangerously unsafe implementations onto the streets because they know their gravy train of funding is about to be cut off. Nobody is ready, none of them are safe. Some are safer under certain conditions, but that is a game of rock, paper, scissors over which people are killed or maimed, not over which company is safe and which isn't.
And plenty more snake oil salesmen are lined up promising if you ad just one more sensor(theirs) to every car then all of this will go away. In reality, the most any of them offer is a small incremental improvement, probably with a dozen new edge cases that need to be handled. So mandating any specific tech is a suckers game, but it's an easy one to sell, and profitable. The real engineering test is to force the cars to operate safely under the conditions they can handle. If a supplier comes up with better tech down the road that can operate under harsher conditions, that can be a competitive selling point, but that shouldn't stop a competing solution from being used in the conditions it can operate safely.
And that's the rub. Despite years and billions in investment, neither the radar or lidar companies can offer a system that can operate on their own. So instead they want to make it illegal to sell a car without their tech onboard. They push easily digestible FUD. But one more sensor is just another version of the "one more fix" problem people mentioned, and their are no quick or easy fixes on offer here. So any system needs to be able to operate safely in it's limits and needs to be able to stay in those limits reliably.
"That is where we need to be going"
Not if it means autonomous cars are doing really stupid things from time to time and getting into accidents very few humans would. While the stats might look good, the perception falls flat on its face. You don't want to have to beg some deity that spilled paint on the motorway isn't going to confuse your car into doing a Thelma & Louise into the valley below.
And humans get into accidents that no autonomous cars will. No robot is going to randomly take 5 seconds to respond to something directly visible in front of them, rather than the normal 0.2 seconds. But humans will….when they are texting, or having an argument with a passenger. No robot is going to occasionally just switch itself off entirely; but humans do fall asleep at the wheel on the motorway. No robot is going to change its route-selection at the last minute, and reverse back along the motorway along the hard-shoulder to get to the exit. We can go on listing human failure modes all day.
The issue is which type of error is more frequent.
While true, you have a person, the driver, to blame in the resulting insurance claim, Police investigation and/or court case depending on the outcome of the incident. We don't yet know what happens when a "robot car" does that. Well, we have a couple of examples where a big corporation has to pull all it's cars off the road and pay a fine/compensation. But there's no one actually to blame or send to prison or ban from driving for a year or three etc.
Additionally, every human is different, some will make mistakes that kill, most won't. Self-driving is effectively thousands, millions of identical drivers all on the road at the same time likely to all make the same mistakes in the same situation.
One of these is a non-issue, the other is just plain wrong.
“The insurance claim” is in fact already fully covered by the system we have today. At least from your personal perspective. Your car, your fault, your 3rd party insurance pays out. The cost of 3rd party insurance premium is simply determined by the accident rate of the self-driving algorithms of that brand. A Tesla might be £500, a BMW £300; but either way that’s for the insurance company actuary to calculate. The free market then determines cars with high-accident-rates to be expensive to insure. And by the way, that’s again absolutely nothing new: a high-end BMW M3 costs a lot more to insure 3rd party only than Toyota Corolla, and that’s not because the car is dangerous but because the driver is.
The bigger issue is that you want vengeance against the car manufacturer, for every case where a human is injured or killed. And that’s exactly why we have laws in our country, not Weregild blood money. As long as fewer lives are lost *on average* self-driving than human, vengeance for individual cases is unproductive and unacceptable.
The issue is which type of error is more frequent
First of all, people won't even consider autonomous cars until their accident rate & fatality rate is 1/10th of the human average, taken across ALL roads and driving conditions. Why 1/10th? Because the average is colored by drunk drivers, texting drivers, sleepy drivers, stressed drivers, bored drivers, and good old fashion bad drivers. The mark to beat isn't the average driver, it is a driver in the top 5% who is sober, rested, and paying full attention. Because I'll bet if you ask people half of them will claim to be in the top 5% of drivers!
People have to be comfortable that regardless of their personal delusions about their own driving ability and focus on that task when performing it, that autonomous cars are still clearly better.
Beyond that, the cars have to very rarely get into the types of accidents that humans never would. Because you're never going to get the general public to accept the risk that in certain conditions their car might drive them into a stopped emergency vehicle on the side of the road at full speed, like Teslas have repeatedly done. Maybe you can never exorcise the demons and there will always be a few corner cases, but they have to amount to a handful of deaths nationwide per year. Otherwise everyone who believes they are in the top 5% and always pay attention while driving will continue to drive themselves, and only be willing to consider an autonomous vehicle if they're drunk - but not for safety (many people believe they are able to drive well when drunk - they'll tell you something like "I'm more focused and always drive slower if I know I'm drunk") but because they want to avoid jail.
It isn't statistics that matter here, it is perception.
“People won’t consider”
Very many groups of people will; you are assuming your own beliefs are held by everybody.
First off, I personally would, along with the set of people who have a science background.
Second, there’s a large tranche among the elderly and disabled, those who want to drive, and for whom driving represents the freedom to leave the house, but are not capable to drive. That’s about 20% of the population.
Third, the young folks do not have the enthusiasm for driving that we do. For them, driving is something that Uber-drivers do, it’s not aspirational.
Fourthly, along those lines, anyone who calls an Uber, and there’s a lot of them, clearly don’t factor safety in at all. There’s a lot of that.
Fifthly, there’s the insurance angle. People can certainly be deluded about their capability, but not when money is involved. Insurance Black boxes are already a thing; for everyone with higher insurance premium due to their record, the deal is going to be “when the self-driving is switched on, you’re earning an insurance discount”. You watch how many people will opt to pay more insurance just to switch the self-drive off in order to do the shopping”
"You don't want to have to beg some deity that spilled paint on the motorway isn't going to confuse your car into doing a Thelma & Louise into the valley below."
Roadworks on the motorway. Lane-assist beeps at me for following the new markings, crossing over the old markings. Two interesting points arose. 1) My lane assist only beeps and leaves me in full control of the steering wheel, no force feedback or anything (good!) and 2) my car managed to detect the "old" lane markings which have been "blacked" over and still managed to detect the change in colour of the road surface despite the temporary ones being new and bright white (road surface was also black, just nor as dark or shiny as the paint on the original markings) - 2) is not so good. Would a self driving car make a different decision based on the new and old road markings? I'd like to think it would see the cones too, but I'd not bet on it after seeing how some self-driving cars handle roadworks.
No idea! It's my first[*] car to have it, it's only about 3 years old and I've not had it long enough to know yet. It two weeks before I noticed the lane assist/warning because I normally drive properly :-)
* previous experience with hire cars, usually for only a day or two, so again, no idea.
I can tell you: it doesn't. This is why this tech annoys me: it is absent and non-functional at the exact moment you'd welcome extra assistance to spot where the lines are so you can keep your position on the road in, for instance, heavy rain.
But no, it only works when you really don't need it.
To some extent, you're correct. A car that is safer than a human can be convincingly argued to be better. However, you have two major problems. People will evaluate these differently because they're a different type of product. If they see them making catastrophic mistakes that they wouldn't have made, they won't be confident in them even if, with accurate results, they're making those mistakes less frequently than humans would. People will freak out over a situation where a human would crash, and that will be hard enough to overcome. You will not help the point by having lots of other examples where a human driver would have been better able to prevent them.
The other problem is that you won't actually know if they are safer than a human driver until you collect a staggering amount of data that is controlled for everything. Set loose ten million self-driving Teslas to drive on large roads, small roads, urban roads, rural roads, at all times of the day or night, in all conditions, then compare the number of deaths caused to human drivers. Then you can use that data in comparison. With existing data where the software is only used on one type of road and with completely uncontrolled and unmonitored conditions, the data is not comparable. Of course, if you try to implement the necessary data collection, you will get some rather indignant public reactions about the experimental not confirmed to be safe things unleashed on the public streets, so you can't actually do what you need to do to collect the data for that comparison. That is why individual situations are considered by road safety regulators before public testing is permitted.
"Set loose ten million self-driving Teslas to drive on large roads, small roads, urban roads, rural roads, at all times of the day or night, in all conditions, then compare the number of deaths caused to human drivers. Then you can use that data in comparison. "
OK, so the data is collected. It would be very hard for Tesla FSD to be worse than the US meatsack average, since the US is comfortably in the top 5 for the civilised world's worst drivers (measured as fatalities per distance).
What then?
Are we to accept that FSD is ready to go, simply because it's not as bad as the meatsacks? Given FSD will likely have a different bias of at-fault accidents according to time, place, conditions, lighting etc, as the evidence shows, are we to say that we'll accept FSD fatalities in low light and poor visibility as a trade off for notably better than meatsack performance in full daylight and on fast roads, say? I don't have any answers here, but if we let perfection be the enemy of the good, then we see more people die on the roads whilst FSD is held back. If we let FSD loose with imperfections, then fewer people will die, but as a general rule each individual casualty that does occur would not have been killed if FSD hadn't been let out of the can. Could be your partner, best friend, daughter who gets mowed down - is that an acceptable outcome of a wider statistical improvement in road safety?
That is what regulators and politicians will eventually have to decide, and even with the data there's no good or inherently right answers.
That is kind of my point. You need to collect that much data, undoubtedly involving many deaths, in order to do a direct comparison in the first place, so you can't do a direct comparison because people won't accept that. Even if you did, people won't follow a purely utilitarian system of "fewer deaths is better, so the method that caused fewer deaths will be chosen". If it's not significantly and provably better, it won't be accepted and people are great at resisting changes to the status quo.
Let's assume that someone, and I'm quite certain it wouldn't be Tesla, made a perfect self-driving system. Even when that system has been put to the test, driving in all conditions, and the only accidents it got into were entirely the fault of something else, for example it was driving along when a massive earthquake struck and toppled it off a hill and into a building, there will be some people who resist its adoption. That will be hard enough to overcome. By painting something as perfect when it's only good, we make that problem worse. By painting something as good when it's actually mediocre, we may kill it before we can get any better. And the levels of quality have to include not only pure statistics but the average driver and pedestrian's view on what is reasonable.
It would be very hard for Tesla FSD to be worse than the US meatsack average
Given that it cannot drive AT ALL in many conditions/situations that's a laughable take. If you did what Musk promised to do nearly a decade ago and have a Tesla drive itself across the country, but disable its ability to hand off to the driver it wouldn't make it out of California before it logged its first crash. Because if it was even close to being able to make it all the way to NYC, Musk would seek out the publicity of the event. But he knows it can't come close to it, and doesn't want the publicity of the event to be people that died near the California/Nevada state line.
But we let millions of poorly sighted people drive on the roads every day
The question is Are the cars causing more accidents than the human equivalents That is where we need to be going
With humans we know who is responsible. A driver who knowingly operates a motor vehicle while he or she has impaired or defective vision has a degree of culpability.
An automated vehicle that continues to operate under conditions where the inputs from its sensors are known to be impaired to the extent that continued operation is hazardous and consequently causes, or is involved in, a serious accident that otherwise might have been avoided - who is responsible?
For Teslas I would like to think Space Karen personally - twenty lashes with the cat would do nicely for starters.
But we let millions of poorly sighted people drive on the roads every day
I don't know where you are, but where I live you get your eyes tested. If you need glasses, it goes on your driving license and you need to provide evidence of recent testing to get it renewed.
Sure, a lot can happen in those 5 years but few are allowed to drive around with a white stick over here :).
I'm sorry but it's pathetic that with these factors:
A) Two decades of essentially free loans
B) extremely permissive authorities that are even more permissive if you tip their elected bosses properly
C) educated and motivated engineers with a common language + culture
D) A frankly idiotic public who will take out a loan to buy any old shit - and who vote in your favored talking head when they get mad at being screwed
That the argument is "but humans do worse!" Come off it.
"But we let millions of poorly sighted people drive on the roads every day"
And we have increasingly strong measures to ensure that they use the corrective measures for those sight problems, and it is from memory an offence to be driving when your sight is below the minimum required standard with penalties ranging from fines, points and loss of licence to jail (if your sight is dangerously bad it can fall under dangerous driving etc). Even going back 30-40 years if you needed to wear glasses to pass your driving test you were told you needed to wear them whilst driving and I believe it is noted on your licence (the 01 after the category).
We as a country are actively tightening up on people driving who have medical issues that make it unsafe, with things like doctors being required to report people who no longer have the eyesight to do so, specifically because of accidents where it turns out people knew someone was driving with poor eyesight that fell well below the safe level and didn't do anything.
It is nuts that we're looking at "driveless" cars and are willing to accept that they're going to be worse than a human for their input, especially if like Tesla's they're aiming to do it with just a camera that can't move, can't adjust for light being too bright, and has blindspots directly in front of the vehicle (meanwhile the average human will automatically make corrections by doing things like adjusting where they look without thinking about it).
How come my PS3 can simulate an entire 3D world track environment plus vehicle and semi-convincing physics, and with its leftover processing power can simultaneously pilot up to 15 competing vehicles? And Tesla, with far greater processing power, struggles to control just one single car?
The PS3 simulation knows exactly where everything is, because it put it there.
That's the difference.
The physics and mechnicals of merely driving are long solved problems.
The challenge of "self driving" is figuring out what is going on, using the limited information available from the sensors on the vehicle, in wildly varying circumstances - and perhaps more importantly, determining what is not currently detectable and (eg) slowing down to match conditions.
Musk then made the task even harder by removing sensors to save money.
" and with its leftover processing power can simultaneously pilot up to 15 competing vehicles?"
It's doing what we would be doing, until those other cars are near enough to matter, we ignore them. We know they are there, but we don't change our behavior. Something like Grad Turismo is great, but I've never seen a "white dog" run across a track. There hasn't been a large rock in the way that has tumbled down the hill side. Water/gravel/mud across the road. If it's been raining, I'm more aware of puddles, water flowing over the road and other hazards. If the temperature has dipped below freezing, I know that there could be icing, especially on bridges. I question whether any of these conditions will be detected by the car and accounted for in how it drives. Debris in the road is always a concern and if there is traffic, the telltale sign is vehicle up ahead swerving abruptly. If I see that, I know there is something in the road and I need to move over, slow down and be ready for it. An autonomous car playing "follow the leader" will be very surprised at the last second and then what does it do?
Game simulation is very poor when compared to the real world. For starters the models are very simple and the physics is also very limited. Most games for example dont really or truely support object decomposition etc. 3d Games are mostly cardboard cutouts with nice textures they are not real or true representations of the real world.
"Game simulation is very poor when compared to the real world."
There are some simulations that are excellent. The cars are well modeled for acceleration, braking, grip and all sorts of other things. The road surfaces are accurate. The problem is they don't model everything. I know the founder of Virtual GT and race car drivers use his stuff to practice for races. Those drivers know it's only good up to a point. Even with some motion, there's no way to simulate the G-forces, wind/aerodynamic effects and many other things. They do get a feel for the geometry of the track and where they will get bounced around and for tracks where there isn't the opportunity to do much practice, virtual is super important.
We have finally got to a point where the value of convenience features is now more than human life.
I don't want to rely on sensors and software written by a commercial company with a budget to demonstrably put my life and the life of others at risk so that I can go "ooh...look....no hands!"
When you get a bit older (sorry), you will view this very differently. Firstly, you will have had the experience, as most people of my age, of having the repeated horrible conversations with your elderly parents of telling them “you are no longer safe to drive”. And Dad says “of course I am” because he lacks insight. Now what? How many blazing rows do you have with an old man? You inform the DVLA? You take his car keys? At this point, he is already accusing your Mum of stealing stuff around the house. And they won’t take taxis, because Mum is afraid of strangers knowing that the house will be empty. When you’ve had this experience, autonomous cars start to look like safety rails.
And as you get older still, you realise that you personally rely on your car to leave the house, and it’s a problem. Getting your disabled and frail wife from sofa into your car in the front drive, becomes a half-hour ordeal, that just doesn’t really work for “calling a taxi”. You can see that “a walk to the shops to clear your head” for yourself, is a simple pleasure that you will be unable to do within a decade or so. So the realisation that *when* you finally become unsafe to drive (rather than if), your life will simply become a housebound 15-year-long Covid lockdown, is a bit of a shocker.
So now, the hope of an autonomous car coming along within a decade looks like a life-saver, for you and others, rather than a convenience. And definitely not “ooh…no hands”.
As a young inexperienced driver, I had my first accident: driving directly into an extremely bright sunset. Well, the guy ahead went round the corner, so I glanced left, saw I could make it too if I treadled it, which I did, then turned to look forward again and saw .. shadow?! Yeah, he couldn’t see into the sun either. Crunk.
(I generally have “ok” eyesight but my brain seems to be constantly looking out for skirts, legs, bosoms, and cops.)
I'm wondering if deep down (or not so deep) Musk knows Tesla are fucked. That's why he's throwing his weight behind the orange one.
Maybe he knows foreign imports need to be taxed, and domestic regulations need to be relaxed if Tesla stand a chance of surviving.
"Maybe he knows foreign imports need to be taxed"
Especially if he can't stay focused and work to catch up. How many models does BYD have now? Nio? I consider the Tesla 3 and Y as one model in two sizes as they are quite similar. I wonder if a designer just used the Liquify tool in Photoshop on the Model 3 and, hey presto, new model.
There's a fine line with tariffs. If the US only jacks up import duties to prevent competition and the domestic makers fail to innovate, the US gets left behind and has to play catch up at some point.
Tesla vehicles are ostensibly just as good at seeing in fog, sun glare or excess airborne dust as a human driver, with the added caveat that it's an AI making the decision, not a human.
I suspect I actually need a cluestick studded with steel nails to hammer the point AI, LLM, computers don't see in any sense that approximates processes the human (or animal) vision implies.
Behind every pair of eyes there is a massive causal model of reality that can envisage what might be in the glare of sun, make a nearly instantaneous risk assessment and decide on a course of action. The decisions might be wrong but Darwin has ensured consistently poor deciders are removed from the gene pool.
The main failing of contemporary AI/LLM is not that the isn't clever or capable but rather that its proponents do not recognise that the human brain and mind is vastly more capable than AI/LLM will ever be - even if the technology were taken to ridiculous extremes of scale.
I wouldn't say that artificial intelligence or sentience is not possible but I can confidently predict not in the lifetime anyone currently living (including Space Karen et al. with pretensions to immortality*.) I would predict that the artiface that our descendants might use, would replicate parts of the biological processes involved in brain function if not in its entirety. Evolution might suggest the path of least resistance from an abiotic/inorganic precursor to consciousness is what we call biological. Fascinating questions that might take centuries to resolve - or never.
* I would be happy to contribute to their barrels of rotten fish. (See Qin Shi Huangdi)
"The decisions might be wrong but Darwin has ensured consistently poor deciders are removed from the gene pool."
I don't think automobiles have been around long enough for Darwin to have had much effect. At best, you might hope there is some cross over from generations of riding horses. But they are still much slower than a car, and the horse itself is adapted to travelling at those speeds.
Meanwhile, by all accounts, Waymo seem to be mostly solving these problems. At the very least, generalising from what Musk's cheap, hacked-together tat can do and what is possible in principle seems a mistake.
" Waymo seem to be mostly solving these problems. At the very least, generalising from what Musk's cheap, hacked-together tat can do and what is possible in principle seems a mistake."
Waymo's approach and that of Elon are different. Waymo vehicles only work in a very defined region that has been extensively surveyed and Tesla is trying to make something that will work anywhere with little supplied data.
My belief is Waymo has the more valid approach for now. Taxis in city centers are more appropriate where manual driving is more cost effective in small towns and on motorways although many motorways would be an easy target for autonomous navigation if there's a good mechanism for detecting debris in the road and other obstructions.
For very long distance travel where it would take several stops in an EV for charging, that's where planes and trains are more appropriate. In the US, both are limited. It's a big country and unless you are traveling from one major city to another, neither are that great. As a rail fan, I can see that before there's a need for HSR, it makes sense to spend money on 125mph class rail. Dedicated passenger tracks, elimination of level crossings, more trains and more optimum scheduling would all be super useful. There's suppose to be a high speed train from Los Angeles to Las Vegas except it won't start/end in Los Angeles but a couple of hours away depending on traffic. I think it would be more viable to have, as much as possible, dedicated, grade-separated passenger tracks built for 125mph trains, not HSR. Given the traffic on Fridays and Sundays, taking a 3 hour train would be fantastic. Knocking it down to 1.5 hours at double/treble the ticket price isn't a big win and extending the route to other places would be out of the question. It would be awesome to be able to take a train from Banff, AB, Canada to LA. A branch could junction in Las Vegas that goes down to Tucson, AZ which is a nice N/S routing that Amtrak doesn't have and HSR is unlikely to be cost effective to build especially considering the international aspect. There's still the option of flying for those with time constraints, but over time it seems that flying for business is getting to be less and less of a thing as comms only get better. OTOH, traveling for pleasure can't be replaced by skype/zoom. Even so, a 6 hour delay on the train that only runs 3x per week isn't going to attract many travelers. Stations that see one train a day at 3am aren't that popular. I'm fine leaving in the wee hours, but I don't want to arrive at that time. Hotels want you out by 10a so that's an issue and small town hotels may not be open for arrivals in the middle of the night.
Yes, Hofstadter explains that seeing is a looping process of observation and comprehension, which explains optical illusions.
With regard to LLM's they are barking up the wrong tree. If the brain used natural language then all languages would be very similar. A point which neurolinguistics has been trying to raise for a long time. Of course the conmen will tell the VC's that symbolic language is what's needed; even Gates says that, but these people would do or say anything for money.
will be Musk's defense. You promising FSD every year since what 2016/17? At least before 2020 , is just madness. He would promise perpetual motion is he could get away with it AND make money.
In that respect, he is just like his master grifter, Donald 'the rug' Trump.
If you are thinking about buying a Tesla please think again. Elon only cares about himself.
Leon, according to his mother Maye, is "Developmently disabled".
According to me, he's a rich idiot that got lucky because of his dad's apartheid emerald mine, and is a evil narcissistic psycopath that's trying to take over the United states, or if you prefer "The World".
The combined power of SpaceX, Starlink and X staying in Elmo's hands is unthinkable if Harris wins in Nov, given his recent behaviour towards UK and US governments.
In addition, letting his half-blind 10 ton autonomous WankPanza on the roads is stupid, and his Robovan looks like my butterdish.
https://www.politicalflare.com/2024/10/elon-musks-mother-freaks-out-online-and-tells-people-to-stop-being-mean-to-her-developmentally-disabled-son/
Early May 2021 Elon Musk revealed he has Asperger's on Saturday Night Live
https://www.bbc.com/news/world-us-canada-57045770
https://www.newsweek.com/snl-read-full-transcript-elon-musks-opening-monologue-saturday-night-live-1589849
This may explain a lot about his decidedly unusual behaviour...
No - there is a massive gap between Asperger's and being a complete d*ckwad.
There are plenty people who score high on this diagnosis and are NOT raging sociopaths who insult people for merely disagreeing with them.
Please do not allow Musk to use that as an excuse for his aberrant behaviour because it isn't.
Dear Reg.
Please update your style guide to eschew the word "accident" for any incident unless it involves a truly unpredictable event like, I dunno, a meteorite strike.
Automobile crashes aren't random. They have root causes. Driver inattention or cognitive overload. Failures in driver perception. Road rage. Poorly maintained equipment or roads. You get the idea.
What is the expected limit of these, at least partially autonomous, systems when related to a human's capabilities?
This relates to comments about Tesla (or others) should be using radar, lidar, ultrasonics etc if they are available rather than visual cues only.
If a human only uses eyes (hearing is very muted in a modern vehicle) should an AI controller be "better" or is "on a par with humans" good enough? Conversely, if an AI has a greater ability to process different data streams and gain better safety performance statistics, do we then ban all those unsafe humans from driving?
Note that neither of these scenarios have the expectation that accidents will not happen, just questioning the safety limits of both systems as it seems to me that an autonomous is not allowed to have an accident but humans are ...
Considering the 'safety' equipment the EU is pushing onto new cars, and how unreliable it is: I'm actually surprised that there aren't a heck of a lot more 'incidents' with Tesla's FSD.
Or perhaps America doesn't have the same problems I encounter daily in the UK... such as the roads that are 90% pothole...
A late addition to this thread for interest only:
Researchers develop system cat's eye-inspired vision for autonomous robotics (sic)
TL;DR: the slit style of pupil in a cat's eye helps cut out glare.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
