Tesla’s Autopilot losing track of devs crashing out of 'leccy car maker Tesla's Autopilot department – developing the software for the 'leccy car maker's autonomous driving systems – has lost a chunky proportion of its staff, according to reports. The exits come in the midst of a wider shake-up of the software unit, according to Telsa fan site Electrek. The site names five engineers from a total …
Autonomous driving is decades away, imho, if we talk driving in a rural Italian village, mere months away if we talk on the highway.
I do think Musk is over-optimistic on that one.
NB: I admire Musk for Tesla and SpaceX, he's done a great job.
hahahahahahahhahahahah
thaks for that, my first genuine LOL moment for quite a while
for me, the FAD bit is never really going to materialise while we 'allow' people to also be on the same roads as the computer driven units, and what about motor cycles, fast as a car, small as a slower push bike, but still carries a very important load, in my case ME :o)
do like the look of the vehicles they are pushing out at moment too, but douby if I will ever be in the market for any car at those prices :o(
> 11 people have left, 5 of whom were named. Roughly 10% of the team = approx 110 people on the team.
> Anything else makes no sense because 5 is not 10% of 11.
It is pretty badly worded.
The Ars Technica article (https://arstechnica.com/cars/2019/07/close-to-10-of-autopilot-software-team-reportedly-quits-after-shakeup/) states that "11 members of the software team, or close to 10% of the total group, including some longtime members, departed in the past few months"
So it could be that the division/team had ~110 people, of which ~20 were engineers (the rest being managers, QA, physical mechanics, etc etc etc). And now it's down to around 100 people, of whom just 10 are engineers.
So a 10% reduction overall, but a 50% reduction in software engineers.
Either way, especially given that Ars hasn't mentioned the 50% figure, the article could do with some clarification!
Yes, but think of the applications to warfare! Perhaps that’s the plan…and how we get to the dystopian future of Mad Max. It’s all brought about by Tesla’s self-driving cars being put on the road too soon. Note those cars that burst into flames also will replace some of the incendiary armaments used in war zones.
"...I see plans within plans…Many machines on Ix. New machines…"
Computer-driven cars only have to be shown to be at least as safe as human drivers to be successful. As you imply, this could be staggered by regulators, with low-hanging fruit becoming licensed by regulators/insurers first and more tricky stuff later.
But then again, for non-natives, rural Italy, India or Africa, are extremely dangerous and the accident rates correlate with traffic density. I seem to remember reading about appalling casualty rates in Kabul's traffic, so that regulators may also choose to prevent some of the worst drivers from getting the behing the wheel.
I think that, while there's plenty to admire with Musk — Tesla has kickstarted electric vehicle and Space X shaken up satellite launching — not all that glisters is gold.
I'm not sure that's tru for two reasons.
Firstly, if I screw up I will get punished (increased insurance and perhaps jail in the worst case) and may lose my license. So, most people will try to avoid accidents although for some it may take one or two to learn. Big companies just hide behind lawyers and will see a few deaths as just the cost of doing business, your brother, wife etc. is dead and there's little to no consequence for the people making money from selling the faulty system.
Secondly, most people try to avoid accidents; So, I realise I'm going too fast but brake too late and hit the car in front. I did however still brake and so reduce the impact. When a computer gets it wrong it is likely that it'll keep the throttle wide open, so even if there are fewer crashes they will be more severe. A certain automatic system in a plane which forced the plane into the ground springs to mind.
This is what's holding everything back.
The human sla is extremely low.
The error factor on teslas sla is likely higher than that figure.
So to make autom deaths + error less than human deaths requires autom deaths to be unrecognisable from zero.
AP teslas are safe enough, its a statistics problem
You don't seem to understand the legal situation very much.
Firstly, if you kill somone in a car accident you could be even be charged with murder: several cases working their way through the German court system like this already and other countries looking on interestingly. However, as to whether the current system is an effective deterrent, I'd be beg to differ. We have more RTAs than ever before but fewer fatalities due to the improved safety features of modern cars.
At least in the US, the principle of unlimited liability and potential class action suits hang over any company offering these services. Which is why they're so keen to get the telemetry to demonstate that their cars are safer. They're also looking for test cases.
It doesn't happen often but companies do occasionally go bankrupt when they put the safety of the customers on the line.
But regulation is also crucial: only idiots will let their cars drive in situations which are not expressly approved, as regulatory approval is their best weapon in a court case. Insurers will also weigh in relying on the data they've been collecting for years to set insurance premiums for computer-driven cars in regulated areas. If the numbers go the righr way. it could soon become prohibitively expensive to insure yourself as a driver.
Your own assessment of your own behaviour and extrapolation for others is a clear case of bias (Kahnemann). We all tend to think we're good drivers when we're only average. My own take is that an awful lot of drivers are already overtaxed by many road situations and increasingly rely on other drivers doing the thinking for them. This is why computer-driven cars are programmed to drive defensively: they know that the other driver is an idiot. Taking a drive in another country, whether it's Italy or Afghanistan, can be a real eye-opener.
The comparison with the Boeing 737 Max also fails here because, as others have explained, Boeing developed a system for the new plane to make it handle as much like the old plane as possible, despite it being mechanically and aerodynamically a completely different beast. The aim was to avoid expensive and lengthy full certification. And now they're grounded and won't get a fasttrack FAA certfication for the rest of the world.
Now, there will no doubt be companies that are prepared to cut corners, sometimes literally, in order to get to market first or cut costs or whatever and there will be casualties as a result. But, at the end of the day, the decision will be about whether there are fewer casualties with self-driving cars as there would have been otherwise because cars stlll kill far too many people.
Alas, there are far too many idiots already behind the wheel. The current emphasis (Nissan commercials on Canadian tv touting their driver assist features) on adding these driver assist functions is only encouraging the new generation's lack of attention to the task of driving. Driving is more demanding now than it was 50 years ago when I learned: much greater volume and density of traffic, moving faster. There is no room left for distractions.
"There is no room left for distractions."
And that includes distractions such as using a hands-free phone. Or talking a lot to passengers.
Research in the late 90s put people in driving simulators on a drive around town, recording the average accident and near miss rate.
Then someone sat behind them and asked questions like Who is the president of America, What's the capital of France. What's 17 - 9, and so on. The increase in accidents and near misses was over 33% !
Okay, it was american research so capital of France was a tricky one...
Humans' brains are single thread processors, no-one can really multitask, we jump back and forth from task to task, concentrating on the one they think is most important. Concentrating on a conversation causes the accidents. More recent research showed that the accident rate in rush hour america (can't remember which cities they studied) was the same before hands-free regulations and a few years after the new regs were put in place were the same.
You've probably seen the lack of multi-tasking when talking to a driver
the words come out okay
and then
......when something
..........distracts them
on...
the road
...
....
.........
their talk
slows......
...up
And then..
goes back to normal when they have solved the driving problem.
Totally ignore your mobile phone when driving, even if you have a hands free device.
Unless your driving a Tesla???
"My own take is that an awful lot of drivers are already overtaxed by many road situations and increasingly rely on other drivers doing the thinking for them. "
Ditto. I've been a passenger in a UK car driven by a middle aged guy who ignored the lane markings at a traffic light and used his basis of "left lane for straight ahead and right for turning right" - on a T intersection with the turn to the left - and then wondered out loud why everyone on the intersection were leaning on their horn at him for blasting straight through and causing them to swerve around him or emergency brake to avoid hitting him.
He'd been driving that intersection for 15-20 years and never bothered looking at the lane markings. I later realised he's functionally blind (both in terms of focus and cataract dazzling) - but refuses to admit it.
Other drivers pull such atrociously inattentive shit that I wonder how they manage to stay on the road and I've demanded to be let out of the vehicle on more than one occasion.
You can criticise younger drivers as much as you want, but they have to pass rigorous tests that older drivers didn't and it shows in the lack of care and attention older drivers pay to what they're doing (there are a few younger drivers who are gung ho, but they're a tiny fraction compared to utterly dipshit 50+ year olds - I'm in the latter age camp and the sheer number of older incompetent drivers _scares_ me. The latest political push is to reduce speed limits on various roads to accomodate these dipshits but that just compounds the problems. They need to be removed from behind steering wheels, not pandered to)
Rabbiting on about driver assist features misses the point that they're not particularly common on cheaper cars which younger drivers can afford. If they're masking incompetence, it's in older, wealthier drivers.
In the last approx. ten years, I've only had near collisions with one type of driver, and she isn't over 50 years old. She's between 30 and 50, driving a Range Rover or similar, exiting our local posh shopping mall with a smart phone in front of her eyes. She makes a right turn on red into traffic, without even a glance to see if there's a car coming in the lane she's about to enter. Terrifying. Infuriating.
I recently had two near misses with idiots both in the big Teslas, which are rare here and, as such, real virtue statements.
But I reckom that if you made anyone drive in rush-hour in a big city they're not familiar with for a couple of hours they'd be begging to stop. The problems and cognitive overload are not that dissimilar to putting a computer-driven car in a new environment, with the distinction that the computer can benefit from the experience of its peers.
Last weekend me and my partner sat in amusement in a cafe watching a Telsa driver totally fail to back out of an angled parking spot, then drive forward into the car in the adjoining bay, then double park and get told off for boxing someone else in, then move forward a few carlengths still double parked next to a narrow bridge blocking cyclists and boxing in another car trying to get out.
Clueless and dangerous doesn't begin to describe it.... The one positive thing you could say is, being an electric car, it all happened very quietly. Apart from the crunching of car against car, and the yells of "Get out of the way!".
Hmm, you want all dipshits in a particular age category removed from behind the wheel although you are in the same age category.
Perhaps the problem, as has been alluded to already is that the majority of drivers believe they are far better than they actually are. This is not necessarily age-related but is a mind set. The is a huge swathe of law abiding people who, once they are behind a steering wheel are convinced that the law is only advisory or does not apply to them:
Speed Limits
Red Lights
Zebra Crossing
Safe braking distances
Assuming that everyone else will get out of the way.
This is also compounded by the increasing number of road users who have no tax/mot/insurance because again, it is cheaper and the likelihood of getting caught now is quite low. What is surprising is that it is not just old bangers and young drivers. Recently the DVLA went on a purge clamping vehicles that were parked on the road that had no tax. There was some flash cars less than a year or two old as well as the old heaps.
You've used the assertion that because of autonomous cars, it'll soon become prohibitively expensive to insure yourself to drive a car manually. I see this assertion a lot, but no-one ever actually explains why.
If safer autonomous cars do start to replace humans it would make driving a less risky proposition for everyone - humans and autonomous cars alike. Care to explain why you think premiums for meatsacks will inevitably go up once the robotaxis arrive?
"Care to explain why you think premiums for meatsacks will inevitably go up once the robotaxis arrive?"
Spreading the risk over a much smaller population means payouts will become a larger proportion of the insurers income, even if the accident rate remains the same or even reduces a bit. Insurers like their profits, Notice how insurance rates charges going up, you can't avoid them because they are a legal requirement, and many of them are pushing black box telemetry units with lower cost insurance (data they can sell)
"Spreading the risk over a much smaller population means payouts will become a larger proportion of the insurers income"
The only way that equates to massively inflating costs piling on the remaining human drivers is if you have them subsidising the insurance risks of the autonomous cars as well as their own.
Even if you assumed the autonomous cars had a perfect driving record, and the accident rate for human drivers stays the same, then the accident rate/risk cost will still scale down proportionate with the number of human drivers. However, as people who are already deemed bad insurance risks will have the most financial incentive to stop driving themselves around, the accident rate is actually likely to drop faster than the numbers of human drivers - in other words you could naively expect premiums to come down. Insurance costs probably wouldn't drop, as insurance companies *really* do like their profits, and you might actually see prices rise a little as economies of scale are lost in a shrinking insurance industry, but that argument won't create a massive price hike for human drivers.
From another poster:
"remaining meatsacks will become riskier and riskier in comparison to the computers"
Another argument that hinges on the autonomous cars being perfect, and the remaining humans would have to cover all the autonomous cars' costs in addition to their own. That's not how the insurance industry works. The autonomous cars will probably be put in a separate risk pool to human drivers (similarly to the way in which females and males used to be considered with separate risk bases) but they're going to have to cover the cost of the risks in that pool themselves. And there will still be risk costs - the human drivers aren't going to be paying out when your autonomous car gets totalled by a collision with a deer, or is torched on your driveway by the local ne'er-do-wells, plus anything arising from near-infallible autonomous cars actually being found at fault(!) for an accident (and if an autonomous car hits another autonomous car, is there an insurance payout? - you betcha there is).
Now, if you said that in a future society with (near) perfectly safe autonomous cars, tolerance for human-caused accidents will be reduced to the point that driving around manually is considered needless risk-taking you might be on to something. In that scenario, human drivers might be expected to pay punitively for all associated damage from accidents, for personal injuries caused as the result of accidents, and for all costs incurred by emergency services (as I understand it the last are mostly just written off as a cost to society in general currently). But that's not because the actual risk has changed, it's because the costs of accidents would have been inflated way beyond what they are today by a society that doesn't tolerate human drivers any more. However, you might as well just say that in such a society humans won't be allowed to drive any more because that activity will be banned - it amounts to the same thing.
Anyway - for this insurance-based driver's apocalypse to play out as planned an absolute prerequisite is the development of autonomous cars with a near perfect safety record *in all driving situations*. Regardless of what his Muskiness says, I'm not holding my breath for that.
The autonomous cars will probably be put in a separate risk pool to human drivers
It's exactly this which is likely to drive premiums for humans, assuming, and it is a big assumption, that computer-driven cars really do have a significantly different risk profile. As the human pool shrinks, it will become harder to spread the risk around but the risk, though, declining will remain higher than the other pool. At some point the pool would start to resemble something like a Lloyds partnership with everyone preparing to take on a share of the payout.
But it's very early days and, unless computer-driven cars do have a significantly lower risk-profile, things won't necessarily change that much that quickly.
human drivers might be expected to pay punitively for all associated damage
It's exactly this which is likely to drive premiums for humans
Not this old fallacious argument again. The insurance premium you pay is proportional to the actual cost of compensation for accidents. There is no reason for human drivers to become significantly safer or more dangerous, so the cost of accidents and insurance premiums will most likely remain in line with what they are today. A free market for insurance will ensure this remains the case.
Actuarial statistics, and assuming the numbers do favour autonomous vehicles (good reasons to think this will be the case, but this is a pure numbers game), then the risk-weighting will change so that remaining meatsacks will become riskier and riskier in comparison to the computers (there will be fewer accidents but an increasing propotion of them will be down to human error), and will hence be asked to pay more, even to the point of being squeezed out of the market if this means fewer overall payouts. Imagine, for example, that there are currently 10 accidents per million km and computers can be shown to have less than 1 per 100 million km, then the insurance companies will have a big incentive (their profit is highest when there are no payouts) to have more computers doing the driving. The telemetry the cars gather will, presumably, also be useful in establishing liability.
There is, in Europe at least, already a bias against humans due to the ECJ banning different premiums for men and women even though the evidence shows that women have more accidents than men, even though these tend to be less serious than the ones men have. As a result of the judgement premiums for men were increased.
But this is just a hypothesis. There are plenty of reasons to expect, er, bumps in the road to a computer-driven approach: just as they can all be expected to get better over time, they're also likely to make the same mistakes. They will also, no doubt, be the target for hackers and criminals.
Not to mention that FSD systems cost money.
Why would I spend an extra $5k$10k,$15k on a car that doesn't drive any better than I can? Sure, it may not make much of a difference to a $100k luxury car, but it's a huge added expense for some $20k runabout that doesn't deliver any significant advantage.
"Computer-driven cars only have to be shown to be at least as safe as human drivers to be successful."
No
They have to be _much_ safer than human drivers
and
Regulators have to decide who is to blame when the car does kill someone.
If Tesla sell FSD as true self driving, well, it is not the passengers, is it?
Who goes to prison for mowing down that child on a bike?
ER, Yes actually. Your assertion is unsupported and a logical fallacy. Though it's common, and is driving a fear of automation to the level of a moral panic. This had already been hashed out in some jurisdictions by liability in the basic road laws.
The owner and operator of a motor vehicle are always liable for damages resulting from the operation of that vehicle. In the event they are not the same person, both liable. Insurance is there to make your life easier and help cover if you can't pay, but the owner and operator still have full responsibility. Adding self drive dosn't change that. If a manufacturer is responsible for a defect they my also be exposed to liability, but that doesn't take it off the operator. If your operating it in self drive, instead of driving yourself, you gave up control, NOT responsibility.
If you think the Auto-drive glitched out, you and your insurance should feel free to try to recover costs, but that's not the same as not being responsible for the decision to operate a vehicle in the first place.
And on that, driver assist features that meet the bar of improving the safety of below average drivers and don't interfere with the ability of skilled but safe operators are just a win for everyone on the road. Holding technology to an arbitrarily high standard serves no useful purpose and will result in preventable damage, injuries, and deaths.
That said, managing peoples expectations as to what can actually be built and operated safely, may reduce the number of people watching dvds while sitting in the passenger seat, or sleeping at the wheel. These systems are not designed to enable people to exhibit maximum incompetence on the roads. Don't climb in the back seat and make a sandwich.
"The owner and operator of a motor vehicle are always liable for damages resulting from the operation of that vehicle. In the event they are not the same person, both liable. Insurance is there to make your life easier and help cover if you can't pay, but the owner and operator still have full responsibility. Adding self drive dosn't change that. If a manufacturer is responsible for a defect they my also be exposed to liability, but that doesn't take it off the operator. If your operating it in self drive, instead of driving yourself, you gave up control, NOT responsibility."
This might be true, but either I'm not at fault when the self-driving car ploughs into a group of people, or I am not getting into a self-driving car. If I am still responsible, I need to be in control, because otherwise that's ridiculous, it's like blaming the passengers for an air disaster.
The reason that a self-driving car needs to be much safer than a human is because we cannot send the algorithm to jail for killing someone. Punishment such as removing licences and jail time both gives an incentive to not drive badly, and also removes the worst offenders from the roads, meaning that the average ability should increase over time, as we weed out the worst drivers. (Of course, new bad drivers are being added to the pool.)
The equivalent would be that if Tesla's algorithm kills someone, that algorithm is deleted, and can never be used anywhere again.
“This might be true, but either I'm not at fault when the self-driving car ploughs into a group of people, or I am not getting into a self-driving car. If I am still responsible, I need to be in control, because otherwise that's ridiculous, it's like blaming the passengers for an air disaster.”
We have to not confuse “responsibility” with “fault”. They can be different, e.g. catastrophic brake failure. If the owner failed to adhere to the maintenance schedule, that’s the owners fault. However, if it was a manufacturing flaw, that’s horrific if it causes injuries, but it isn’t your fault and you don’t go to jail. You aren’t expected to personally vouch for every piece of metal.
But you are always expected to take legal “responsibility”. Simply the third party insurance pays out, and that’s why we are legally required to have it. This takes civil legal responsibility, separating it from criminal fault.
Your insurer has the actuarial records of safety of the *car*, and agrees to take the responsibility risk. In practise, drivers are higher risk than cars, but car failures *do* happen. E.g. Toyota and their uncommanded acceleration flaw. Self driving algorithm is identical principle.
The car manufacturer may or may not get hauled over the coals for manufacturing flaw. Even a traceable manufacturing fault doesn’t make the car manufacturer go to jail. It depends on negligence; sufficient testing of an algorithm can disprove that. The Uber case, that was criminal negligence on Uber’s behalf. Only in the U.S. will Uber get away with that. I agree that I wouldn’t want to get into a self driving vehicle in the U.S., but then I wouldn’t travel there anyway.
"The owner and operator of a motor vehicle are always liable for damages resulting from the operation of that vehicle."
Wrong. The owner may be held partially responsible if the vehicle is not properly maintained and that can be shown to be contributory to the accident, but in general, the driver, if the cause, is the only one held accountable. If I let someone else drive my car in the knowledge that they are properly licenced and insured, then why the hell should I be responsible for them doing something stupid and causing and accident/injury/death? And going back to the vehicle maintenance, it's the driver who is responsible to make sure the car is safe and legal to drive before setting off, even if they are driving their employers vehicle and are threatened with being sacked for not driving it.
If I let someone else drive my car in the knowledge that they are properly licenced and insured, then why the hell should I be responsible for them doing something stupid and causing and accident/injury/death?
Agreed, but this is where the issues lie.
Let's take a hypothetical situation on here which could actually happen today: Let's say a company purchases an automated piece of machinery. They use it in accordance with the manufacturers guidelines, but someone is killed by it (who also did nothing against company policy or manufacturer guidelines). The systems in the machinery had encountered something the manufacturer never expected, or had failed in some way, and this had led to a death. Who would be liable?
In my mind, the manufacturer should be liable in this case. If the company or victim had somehow violated the guidelines or done something wrong, then they should be liable, but if they haven't then the machine killed on it's own, operating by the instructions given by the manufacturer, therefore they (should*) be liable by common sense.
The same should be true for autonomous vehicles: as long as the owner and/or "operator" (/passenger) of the vehicle has followed the guidelines of the manufacturer (installed updates, regular maintenance, correct proceedures etc) then they shouldn't be held liable for any damages cause by the vehicle doing it's own thing. The manufacturer of the vehicle should assume such liability.
* If anyone knows the actual legal position in the hypothetical above, please let me know. My reasoning seems correct to me, but I know that the law may be completely different and would be interested to know how that would play out.
"* If anyone knows the actual legal position in the hypothetical above, please let me know. My reasoning seems correct to me, but I know that the law may be completely different and would be interested to know how that would play out."
You have obviously never experienced the thrill of being the target of a horde of ambulance chasing lawyers. My experience is the "actual legal position" is dependent on what the presiding judge will allow into the lawsuit.
My example isn't automotive but it is relatable: I once was employed as a repair technician in a General Aviation Avionics Repair Station. One day, a small Cessna was flipped inverted by the wake turbulence of a departing airliner and contacted the runway with its wheels pointing skyward.
Our company was named a co-defendant in the lawsuit because we were listed in the maintenance records as having performed a repair on the aircraft - over 12 months prior to the incident we had replaced the loudspeaker for the navigation/communication radio.
Here is the way the ambulance chasers work here in the States: If your (company's) name can in any way be associated with the accident, you're going to have to spend time and money defending yourself against scurrilous and logically unfounded charges regardless of what the laws and regulations might say.
This is exactly why the self-driving cars need to be vastly more safe than the average driver before they'll be viable at all.
If a human being causes an accident we expect them to be accountable, either through insurance or in court, depending on what's appropriate. Making the human owner (likely a passenger or not even present) legally liable in full would be a step too far for most people to accept. The moment an otherwise innocent owner of an autonomous vehicle takes the fall for their pride and joy committing vehicular manslaughter is the moment that the self-driving industry dies completely. If they were only on a par with humans for safety autonomous cars would be dead and gone within a month of launch - no-one would get in one, let alone buy one. With this liability model the robotaxis will need to be vastly safer than humans to last long enough to become a fixture in the car market.
End users/owners of cars will always be liabile for any accident caused by maintenance issues, but in reality the liability for driving decisions made by the car itself can't be completely dumped on the end user. Any product sold in the UK* has to be "fit for purpose". Any product found to have design flaws that renders them unsafe needs to be recalled for repair, replacement or refund - and any damages caused by such a fault are be a liability of the manufacturer. Every single accident caused by a decision made by an autonomous car could reasonably be challenged under the "fitness for purpose" regulations - so the liability for unsafe decisions could well end up back with the manufacturer regardless of what their EULA agreement says. For the most serious caes it won't be possibly to imprison these companies for causing death by negligent software, so extremely large financial penalties will be used instead - which will severely mess with the economics of making such a product if the cars were only roughly as safe as a human driver. Once again, It looks to me that the robotaxis will need to be vastly safer than humans before the economic model would make sense.
* most other countries probably have similar regulations
I for one would be very interested to see how Full Self Driving performs in 3rd world nations.
Between the 5 cars abreast in 2 lanes, the constant horn sounds acting as a form of radar, the poor infrastructure - roads as well as lack of signage/lines, goats and chickens (and children) running around, and of course the general lack of driving laws and/or observation of same - would the FSD vehicles even be able to move?
Of course, there are exceptions. The dictators and billionaires will have a police escort...
Never mind 5 cars to two lanes, in parts of North Africa I visited using the correct side of the road was optional, including the full coach of people (inc me) that did a U-turn through the breaks in a motorway barrier, and drove the wrong way down the wrong side for a bit, to get to the nearest fuel station.
At night. He had circa 50 people on board, and we followed a string of lorries doing the same thing.
"regulators may also choose to prevent some of the worst drivers from getting the behing the wheel."
This in spades. As soon as a computer can approximately match a half-decent human driver most humans will let them do it - which is a good thing because a half decent computer will stay half decent all the time. vs even a good human spectacularly losing concentration at least once every 5 minutes.
It actually takes a pretty bad series of errors to cause a crash (not all of them made by the driver - roading "designers" are frequently unqualified utter numpties, particularly in places like the UK where road speed limits and layouts on major roads are not signed off by qualified engineers but by local politicians operating on a knee-jerk response - and such numpties frequently engage in victim-blaming instead of sorting out shitty designs that kill people.)
Insurance actuaries are going to be the deciding factor. It's been known for 60 years that drivers who travel substantially slower than the herd have a significantly higher crash incident and claim rate(*) than the norm whilst those who travel substantially faster than the herd have a higher per-claim level(**). People who drive too slow when they have black boxes fitted may well be putting their premiums UP...
As soon as robocars can show statistically lower crash rates in various scenarios(***) I can see premiums for manual control skyrocketing and/or insurance companies insisting on enhanced driving training before they offer coverage at all - whilst much higher standards of driving competence will eventually be demanded by regulators simply because robots will set the minimum level to be reached..
(*) The 2 most common insurance claim categories are "backed into another car whilst parking" and "my parked car was damaged by a hit and run driver" - usually in parking lots. Going slower doesn't make up for poor spatial judgement, rotten eyesight, etc. The next few categories involve changing lanes into another vehicle without checking or low speed scraping walls, poles, curbs etc.
(**) higher speeds == higher crash energies == bigger mess.
(***) Particularly highways/urban/suburban highways. There will always be situations where the robots can't cope but the response without an advanced driving certificate is likely to be to either allow manual driving at a low speed or to switch to a remote handler - and you can pretty much guarantee that cloud learning will be applied. Whilst Uber spectacularly fucked up their programming with assumptions based on politically motivated traffic laws (instead of safety-rooted ones), Google's done quite well at hazard anticipation and handling (particularly pedestrians and other items on the roadway)
"I can see premiums for manual control skyrocketing and/or insurance companies insisting on enhanced driving training before they offer coverage at all"
Why? My risk hasn't changed. So I'm still a £200/year risk, whether terrible drivers are priced off the roads or not? Now, if you mean to say that insurance companies look for any excuse to rake in more cash, and this will be a good excuse to increase my premiums then that is true, but that would require them to all form an illegal cartel to do that.
I would note that the auto-driving cars have a counterpart to driving while impaired:
snow, fog, sandstorms, dirt on lenses, IT failure, laser pointers, really bright reflections of sunlight, the list is endless of the ways by which the sensors feeding the self driving AI can be affected.
And this ignores the computer vision shenanigans that pranksters/thieves/malicious could do - making stop signs look like speed limits and what not, or data poisoning the maps.
The point isn't that any one or dozen scenarios can be trained against - the point is that there are probably an infinite number of such scenarios. How will the self driving machines handle the ones which it isn't trained for?
They need to be shown to be about 10x better than human drivers. For one thing, humans have a wide range of driving abilities, but most of them think they are better than average. Would YOU let a car drive if you it was known to be "exactly as good as the average driver"?
Second, people have a lot more tolerance for mistakes they make themselves versus mistakes that are not in their control. Look at how upset people get over plane crashes despite air travel being far safer than car travel overall.
Third, people drive in ALL conditions, and until cars drive in ALL conditions they will be self selecting for the easiest driving like expressways, and avoiding the difficult driving like in a snowstorm. So metrics like the bullshit Tesla fanboys try to push claiming Autopilot is already safer than humans are a crock. The comparisons must be like for like.
"Second, people have a lot more tolerance for mistakes they make themselves versus mistakes that are not in their control. Look at how upset people get over plane crashes despite air travel being far safer than car travel overall."
The same seems to apply to coach ad bus crashes. It;s probably because dead and injured were passengers, not in control, and the numbers involved in a single incident. Car accidents generally involve few people and of those few people, a significant number are the people in control, the drivers. The human condition is a funny old thing.
The obvious step would appear to be for Tesla to test vehicles in those very places. Are they doing that?
Actually, I'd be interested in how it manages round the country lanes in my neck of the woods. Yes, have seen a few on the more main roads, but not yet on the really tough ones. And no idea what mode they were in.
Plenty of roads round my way that are single lane with passing places. Sometimes you end up having to reverse to the last one you passed to let an oncoming stream of traffic get by. If someone comes along behind you whilst you are doing that it makes it ever more complex, especially if you're reversing round a bend.
Sometimes you have to be practical to resolve situations, even if it means increasing the risk (which means you take things easy to minimize it).
A risk averse algorithm would just end up stationary in the middle of nowhere, along with everyone else using that bit of road.
I've driven those one lane roads in Ireland. At least I didn't have to worry about driving on the left :)
There aren't any around here, but I've driven on some in Nebraska. They are not uncommon rural areas where paving two lanes isn't worth it where traffic is a few dozen cars per day.
We'll know true (i.e. level 5) self driving cars are about ready for prime time when they can handle these without breaking a sweat. When they can handle driving on them when they have a couple inches on snow covering them and no tracks to follow, then we'll know they ARE ready.
Computer-driven cars only have to be shown to be at least as safe as human drivers to be successful.
That depends entirely on what you mean by "successful".
If you mean the number of deaths/injuries on the roads has reduced, and objective measure of the success of the system, you'd be almost right. It would need to be, at a minimum, ever so slightly better than human drivers. That aside, I'll agree on that point.
However, if you mean "successful" in terms of the number of people trusting the system, uptake of self driving vehicles etc, they will need to be safer by a large margin. Think 100s, or even thousands, of times safer. The public is currently numb to the number of accidents on the roads but will not be numb to a "robot" causing those accidents. An RTC can be blamed on a "stupid driver" (not all drivers), where a collision caused by an autonomous vehicle will cause doubts in all self-driving cars. You can see it already: Every serious incident involving a Tesla on Autopilot is reported, even though (from what I've read) in terms of accidents per X miles there are less than for human drivers.
Look at it this way: There were 1770 deaths on the UK roads in the year ending June 2018. If 1% of people switched to autonomous cars and their safety level was the same, this would be 18 deaths in self-driving cars in a year, or more than one every 3 weeks. If people were being told of a death from autonomous cars every few weeks on the news, they would not be likely to buy or use them, even though 100 people had died in normal cars in the same 3 weeks. If you included serious injuries, that goes up to 27 per year or more than 1 per fortnight...
This will only ever work when ALL vehicles on any given road are linked up via a fully autonomous traffic management computer system that has complete control over every single vehicle on that road.
And I for one welcome our autonomous traffic management overlords.
…..and every pedestrian, dog, cat, bicycle et all transmits it's real time location down to the nearest foot ….. yea right …. have you guys watched people in 'real' driving situations trying auto pilot? While it may work on well marked/delineated highways, on your everyday rural/urban roads it can't cope and it's going to kill more people (and no, you can't just blame the driver for trusting it, it's being miss sold and over hyped)….
Yup, both the Mark 1 Eyesight from the S, and the Mark 2/3 in the 3 and newer Teslas, as well as 2 of the newer but crappier systems on other non Tesla cars. I am actually surprised by how well the Tesla system handles bad and non existent lane markings. It is also "cheating" and following the cars in front of it though. Just like human drivers, that's great until the car ahead of you wipes you off on a barrier at the last second by swerving.
I also remember both of them being pretty finicky about even ENGAGING autopilot on secondary roads, and kicking out at the drop of a hat. And yeah, autopilot is a choice, you turn it on, you own the consequences. It's not blame, it's responsibility. Buy the ticket, take the ride.
Autonomous driving is decades away, imho, if we talk driving in a rural Italian village,
Based on my experience of Italian drivers, I don't think it would take that long to develop an algorithm for just following the defined road line while sounding the horn and paying no regard for other road users.
Ladies and gentlemen, we have reached a major milestone on the way to FSD vehicles. We've reached FAD - Fully Assholish Driving - a necessary step on the journey. It will have to be customized for local driving characteristics, of course. For example, in France, all sensors on the left and front of the car will be ignored, since the only known rule is "yield to the right". We're doubling our R&D team just to handle Boston because no one knows HOW they do it there. Stay tuned for local updates from your nearest dealer.
I see putting an autonomous system onto a Italian or Greek road as cruel and unusual punishment.
You are right, its many years away. If a smart 12 year old cannot get a license why would an autonomous car which is far far more stupid.
"If a smart 12 year old cannot get a license why would an autonomous car which is far far more stupid."
The problem is not intelligence. The problem is emotional driving.
Humans are dangerous because they drive angry, drunk and distracted, or simply well beyond their abilities, use vehicles as weapons and as penis extensions.
The easiest proof of this is the way people race off at green lights - a green light is not a facing "GO" signal, it's an indication you may proceed if and only IF the way is clear to do so and you have checked nothing is coming the other way. Drivers and cyclists forcing their way through pedestrians still crossing are amongst the most dangerous things on the road imaginable.
The _vast_ majority of people objecting to robocars are doing so on the basis of feeling emasculated, not on actual safety grounds. They're the ones who shouldn't be driving in the first place. (It's the same issue as power attracts the corruptible and those who seek it should never be allowed to have it)
Not true, I would LOVE to hand over the driving to my car. I could go out and get drunk and have it take me home. I could have it drive 1000 miles to the east coast while I slept, surfed the net and watched movies instead of getting groped by the TSA and taking about as long to get there if I count up all the waiting around and travel to/from the airport. Errands to the grocery store could go either way but I certainly wouldn't feel "emasculated" by having the car drive me. Anyway, it seems like that would only be a potential problem for half the population anyway.
I get pleasure from driving maybe 2% of the time I drive, but I'd be happy to sacrifice that if I got a car without a steering wheel that never needed one. But the fact that humans like me can get confused sometimes when signage or construction areas aren't clear doesn't give me much confidence that software will be able to handle it until I'm so old I'll have no choice but to hand over my driving to software.
Getting 99% or even 99.99% of the driving is the easy part. Its that last bit of really tough/confusing/etc. that will take forever and will result in MORE accidents whether you say "fuck it" and let the car try to figure out and accept some deaths or the car wakes you up and says "we've reached a place where I don't know what to do, so you will have to take over if you want to go any further" and people who hardly ever drive (or in the case of younger people, have almost NEVER driven) are suddenly saddled with the most difficult driving situations possible. Yeah, that's a recipe for success...
"drive 1000 miles to the east coast while I slept, surfed the net and watched movies instead of getting groped by the TSA"
That raises an interesting point. Could a successful genuine self driving car kill off the short-haul city hopper flights/train journeys because, as you say, travel to and from the terminals plus, in the case of flying, the delays caused by security theatre make the journey time so long.
I probably fall into the race away from the lights category but i focus on situational awareness and i believe a slightly more "exciting" drive keeps me focused. 25 accident free years in a car (not on a M/C unfortunately as people really don't see you).
If you drive like you are sitting on your home sofa then you are quiet likely to switch off and miss clues to situations that are developing.
For example on the motorway i can often tell when people are about to pull into my lane by keeping an eye on how fast they are coming up on the car/truck in front of them, they don't need to indicate and often don't but i will already have moved out of their way just in case.
If computers can get to that level of anticipation then that will be a major safety leap forward.
Autonomous driving is decades away, imho, if we talk driving in a rural Italian village, mere months away if we talk on the highway.
I don't think it's coming, at least not to ancient and populous capital cities like London.
Most cars & commercial vehicles follow most rules most of the time, and that's about as good as it gets. At the other end of the predictability spectrum are cyclists and toddlers who generally do whatever they feel like in the moment with scant knowledge of what the rules actually are, much less an intent to obey them.
Absent strong predictability, a self driving car has absolutely no chance of getting things right most of the time. Banning cyclists and putting railings up to deal with dogs/toddlers entering the road randomly would be more politically difficult than simply not approving the use of self driving cars.
Politics rather than technology is what will hold back autonomous vehicles over the coming decades.
The site names five engineers from a total of 11 who have left the company.
Does this mean Tesla is developing FSD with a team of 11 (now reduced to 6)? With due deference to The Mythical Man-Month, that seems a bit thin.
[edit] I suppose it might mean that 11 engineers, of whom 6 remain nameless, have left.
This post has been deleted by its author
Hmm - trouble understanding the numbers referred to in the subtitle here - only 11 developing the software? If there were, they won't have problems finding other jobs, phenomenal productivity, even given the SNAFUs.
But, Musk's manglement philosophy does appear to be one of "the beatings will continue until morale improves"
As someone who lives in a rural Italian village I think you're being a bit optimistic there. Apart from the usual problems It'll have to cope with people driving on the wrong side of the road (just because), overtaking on blind bends or hills (or both at the same time while talking on the phone). Randomly stopping in the middle of the road, speeding up and slowing down for no reason, and that's just what I saw yesterday.
NB: I think Musk is a repulsive gobshite.
Then there is the difficulty of programming different driving modes - mode 1: driving like a native to the RIV (driving on wrong side of road etc) and mode 2: driving like a visitor (reversing for miles when bus/garbage truck/local is coming other way and refuses to give way).
The same issue exists particularly in the UK west country. I live in rural Cornwall and just about every village is accessible via single track lanes with what are euphemistically called passing areas (just about wide enough for 2 medium sized vehicles to go through with care).
This can be interesting as some of the locals and in particular delivery drivers go down these lanes as fast as possible (speed limit 60!) oblivious to the thought something may be coming the other way.
Now, if an oil truck (a lot of homes here have oil fired Agas and Rayburns) is coming the other way - well, back up to a decent space which might be several hundred yards.
Saw this on a bumper sticker a while back: "Welcome to Cornwall. your car is not as wide as you think".
It's a pipe-dream, ain't gonna happen.
The people who have this thing enabled on their cars are just signing a suicide pact that involves other driver's unwittingly.
Even with a "FSD certification" from a government, I'm still not gonna trust it.
Am I a luddite? No. I'm a realist. With a sense of security and what computers are actually capable of. Voice/face recognition was pathetic in the 80's, it's still pathetic now. That's because it's a *hard* problem to describe to a computer, nothing to do with the power you put behind it. Self-driving is exactly the same.
Google's new Go engine was the biggest, most radical, most humungous step forward in AI since it was invented. Going from the best machines in the world barely beating an amateur to thrashing all the masters, almost overnight. It's unbelievable, I was *so* pleased to see it, I imagined it was some amazing new way of doing things - because I studied Game Theory, Graph Theory and various computer science courses to degree level, and one of my tutors was one of the world experts at making computers play Go. But it wasn't. Little else has resulted from that. It's a very limited niche that it made a leap in and isn't translatable to more complex AI problems.
Musk just has no concept of what he's asking, and what he's selling, and what's actually possible. You'll end up, at best, with a poor AI-driven thing that'll be involved in just as many crashes but in all the "less obvious" scenarios. Sure, it won't fall asleep on a long boring straight road, but it's going to plough down a police officer who's trying to move traffic out of an obstructed lane on a motorway without any special road markings to do so.
And I've said a million times before - stop testing *on the road*. There are a billion test cases you could use to build confidence in such a system that don't risk multiple serious deaths. Have an AI-controlled bus punting around Disneyworld. Make a self-driving shopping trolley, or golf cart (how easy can you get - 18 holes on a closed off course that you can layer all kinds of signals over and minimal hazards), or fairground ride, or airport vehicle... all low-speed, less-impact tasks in controlled, closed-off areas where you can prove that it, say, never even bumped a human in three years of operation. They could have started that TEN YEARS ago and got that reputation and *then* translated it to the real road.
But no, Musk just goes "70mph on the motorway only and tell people they were wrong if they activate it anywhere else!". Idiot.
While I agree with some of what you said, I think it's simply more likely that Musk announced what had to happen for Tesla to not fail as a business.
You will eventually have to do tests on the road, as it's stupidly vain to think you can anticipate all possible issues. Heck, we can't even get Raspberry Pi USB3 right.
Voice recog is actually pretty good now. The part of transcribing a particular person's utterances has long been solved. The only addition (other than snooping) that having it done on a bigger AI and server farm is adapting to *any* voice, when the issue is, it's easier to tell which human is speaking (voiceprint type data) than what is said. And that's only peripherally relevant to auto guidance, where a host of other problems arise.
I'll agree all that crap called AI has serious issues. I was involved in the previous wave, which predicted the kinds of errors one would have if one simplified the activation function, and added layers and width - correctly. And we see it was a good prediction, though you get plenty of pushback from those selling the current overfitted models.
Um, it's not that a problem is "hard to describe to a computer" with neural nets. That's not how things work....you're thinking conventional programming or fuzzy, rule based logic perhaps?
But in the end, I agree if for different reasons. Lives are at stake, humans aren't perfect either, but make different classes of errors, and with humans the liability is much more sorted out. We haven't solved various interesting liability problems for simple stuff involving software yet (Microsoft would be dead broke....) - this is a hard one on more than just the sensor (you left that out) and computer level.
I like driving, myself...so I think this is all foolishness. Some people shouldn't be behind a wheel, and giving them this thing will only make it worse for the rest of us for quite some time. Although I think it might be easy to be sure a computer won't overtake on a blind hill while on the phone...
While I agree with some of what you said, I think it's simply more likely that Musk announced what had to happen for Tesla to not fail as a business.
That's certainly a possibility.
Heck, we can't even get Raspberry Pi USB3 right.
They key difference being that a Raspberry Pi that does almost everything right is useful for most folks whereas a Tesla that does almost everything right is a menace.
"stop testing *on the road*."
You have to test on the road - the point is to make it iterative and to make the introduction gradual.
My 18-year old car has adaptive cruise control which only works down to 20mph and 30% braking. It was only after experience and confidence with technology that makers were confident enough to allow it to apply harder braking and let it run down to a full stop/start setup.
We've been talking about robots taking over for decades. They already have in many areas - when was the last time you saw a room for of ledger clerks scratching away? or (more recently) keypunch operators? It's a gradual process, Tanks don't "suddenly get parked on lawns" unless someone's been holding out for years, refusing to adapt in the face of constant change (like British shipyards did) and finally find that they HAVE to do it all at once or go out of business.
Thank you!
Even the homicidal clown engineers at Uber had done(perhaps also stolen in Uber's case) years of testing off the public roads. There comes a point where you have to get in the real world to continue to improve the systems performance. An it is unlikely that ANY system only tested "in the lab" will be safe in the real world.
The idea self driving cars have to emerge from the lab without extensive road testing is a great way to kill people.
We will all share a bit of risk while we sort this out, but we survived the introduction of cruise control and the automatic transmission too. So far, the autopilot systems have a lower body count then Segway scooters and hoverboards.
Where I live, human drives learn on the road (under supervision) but aren't allowed on the motorways.
Self-driving cars (well, Tesla autopilot anyway) seems to have been developed to drive on the motorways first.
I think the justification for the former is that although motorway driving is often only challenging because of how tedious it is, if you do get into a dangerous situation it's going to get very, very difficult indeed with no notice at all. Learners are understandably kept away from those situations until they've proved that they can handle a car in slow moving traffic, together with all the distractions that come with driving around pedestrians, bicycles, and animals on smaller roads.
Auto-pilot has been unleashed on motorways first because for the majority of the time that's where the easiest, most routine driving is. It gives Tesla something to show to their investors, despite it mostly being not much more than an adaptive cruise control system. Of course, when things get sticky (lane ends abruptly? Car in front the auto pilot was following swerves suddenly? Semi truck parked sideways across the carriageway ahead?) then the results are pretty dire.
Now, I'm sure FSD capability will probably have a much enhanced degree of awareness of the traffic around it, but something tells me that once again it's going to be deployed on the faster (usually more routine) roads first, and will shut itself off on slower/busier/more confusing roads at the drop of a hat (either that or it will show itself up badly in unusual road situations). It'll also end up causing some really nasty accidents on the faster roads, when things get a bit sticky and the human can't take over at a split second's notice. It'll still have much better safety stats than human beings for the routine motorway stuff, but that's only ever been the most important metric if you're a Tesla investor.
How well does the other technology you experience work? You never get a phone call from some exacerbated relative pleading with you to explain how to get TV + sound at the same time? I’ve been in elevators that are terrifyingly unable to understand button presses; thus autonomously failing on the most restrictive closed track you can imagine.
On the other hand, during my last few visits to Palo Alto, I notice far fewer jay-walkers than before....
Not sure how people have become confused.
The article says that there is a 10% loss of workforce (i.e. decimated), with 11 people leaving of which 6 (of the 11) were named in another article.
So eleven have left representing 10% workforce, or an initial workforce of 110 people.
That's still quite a sizable team - especially as managing software engineers is akin to herding cats.
No it doesn't. It says right at the beginning of the article:
Tesla's Autopilot department... has lost about 10 per cent of its staff... The site names five engineers from a total of 11 who have left the company.
It's quite clear that 11 engineers have left the company, 5 of which were named by the site, and those 11 make up approx 10% of the staff in the dept (although whether this is 10% of the engineers or 10% of total staff, and if the latter how many of the staff are engineers, is in doubt).
If you think this is badly written and didn't understand it, you obviously haven't spent much time deciphering user/client specs... This is absolutely crystal clear by comparison!
I parsed it as 10% of the division's overall staff, including those 11 engineers. Not that those 11 made up 10% of the entire division.
But then again, considering Musk has taken direct control of that part of the company recently, there's no telling how many less prominent people have either walked away or were pushed out as a result. Turnover rates in some parts of Tesla appear to be comparable to that of a fast food restaurant.
The problem with insisting on "changing the world" is that one must then demand "change the world" class performance from one's subordinates.
And despite all the hype, it should be obvious that there are only a few who can qualify for that level of insight and implementation skill.
So if you're serious about it - you're going to be quite lucky to find even a few such people. Many will be called, but there will be lots of shakeouts - particularly the ones who aren't total hypocrites when they discover that despite the BS on their CV, they really aren't that good.
It should be obvious that the reason some of these problems weren't solved is that they are either extremely difficult to solve even with pretty serious resources - or won't be solved under current abilities to define likely solutions.
I mean, that's why it's a big deal when someone does solve them.
A few key things noted as missing from their latest FSD demo: detecting speed limits, navigating roundabouts. A few things not right with use in the UK: Zebra crossing zig-zags causing steering wobble, high contrast shadows under bridges causing phantom brake, through/right lane road arrow markings causing dangerous unwanted manoeuvres off navigation route, coping with average 2.4 m parking bays (US ones no narrower than 2.7 m typically). Plus it doesn't help that the EU cripple all systems from auto-steering tight rural road bends.
Elon Musk seems to share Donald Trump's problems with tweet-continence. Again it looks like he's overpromised.
They will get there eventually, but what's clearly needed is more baby steps. Maybe the next one is safely disengaging from autonomous driving even if the driver is asleep or unwell - or drunk! This would obviously sometimes involve parking after an assessment of risks. Once this is in the bag, they can take on the easiest roads first, like clear or slow-running motorways, and then disengage as predicted conditions get more risky.
Remember that the Autopilot that's already installed can't be bad as the collisions we've heard about are fairly low when viewed besides the number of miles driven with Autopilot engaged. Remember also that, despite what people may say, we don't need perfection - we do need the autonomous driving to be demonstrably safer than human driving.
It can be done. Maybe not in a year, but not too many years. I'm confident that I will be relieved from the drudgery of motorway tailback driving within the next five years. And that's one of the most important things we could most of us do without.
Remember that the Autopilot that's already installed can't be bad as the collisions we've heard about are fairly low when viewed besides the number of miles driven with Autopilot engaged.
The self-driving feature on any AI car tends to only be enabled in optimal circumstances. Human accident statistics, on the other hand, include everything from divided highways on a sunny day to blind turns with driving rain. The two aren't as comparable as advocates like to pretend.
"The self-driving feature on any AI car tends to only be enabled in optimal circumstances."
You mostly mean "boring ones" - which are the ones where humans lose attention, do stupid things and crash.
The harsh reality is that despite the hating - and despite the high profile twats taking the piss by changing seats or sleeping in the driving seat - crash rates are statistically lower on cars with driver assist features and on Teslas quite noticeably so. They ARE keeping people out of trouble.
They're not robocars _YET_ and they won't be for a while, but they're improving. In the meantime they need a cattle prod in the seat cushion to ensure the driver is awake and watching what's going on.
The _long_ term outlook for robocars is a sharp _reduction_ in the numbers of vehicles on the road, because people who don't feel they need to drive don't feel they need to own a vehicle and are more likely to use a hire vehicle - as the most expensive part of a hire vehicle is a driver, robocars lower the cost of hiring dramatically AND they encourage the use of public transport over longer distances because it makes hire vehicles more likely to be found even if you need to go to some remote spot in the middle of nowhere (a robot doesn't have to go home to sleep, or have 12 hour limits on driving).
The effects of robocars is even more marked in HGV safety - given that the vast majority of HGV crashes are down to tired drivers reacting too slowly or impatient drivers pulling something stupid on highways, automating this part is likely to have a big knock on effect - especially as it's hard to recruit new drivers in most countries.
"as the most expensive part of a hire vehicle is a driver"
No it isn't - not even close. Cab drivers make peanuts.
And that's why this 'no one owns cars and everything's automated' vision of the future seems a bit off to me. We already have self driving cars, they are called taxis. People still own their own cars, because ultimately it's WAY cheaper than catching taxis everywhere.
Removing the cab driver and replacing it with an AI that took tens of billions to develop doesn't seem a dramatically cheaper way out to me, even if they do ultimately make it work.
Even the most powerful pattern-recognition engine in the world, aka the human brain, frequently makes mistakes when driving and these mistakes are sometimes bad enough to kill people. How can an AI whose substrate can fit into a car even come close to that? The only way FSD is going to work is if all cars are interconnected and the roads themselves are intelligent enough to guide them.
No, the only way it'll work is if you remove all intelligence from the system.
This is a road. There's a BIG LINE down the middle. Encoded down that line is all the information you need about the road you're on and the road ahead (in case the line fades out!). When a car deviates off the line, it stops. The "self-driving" cars literally just rely on being told EXACTLY what to do, when. And tell each other exactly what to do, and when. I am 5m ahead of you, slow down.
The drive to put "intelligence" that we don't have and can't even define into things that are not, and may never be able to be, intelligent is just stupid. What you want is artificial stupidity - computers that obey orders, perfectly, every time.
You wanna change lanes? You have to wait for a lane-change line to approach, then announce that lane-change to everyone around you, and then follow the line you're given.
Such things only work on a "dumb" road. With other "dumb" cars. And "dumb" junctions. Make life easy for the computer. I don't understand why we're deliberately trying to make life difficult for the thing we're putting in charge of ours and other's lives. Make it easy. The same way that we designed a steering wheel for humans to steer the car easily. So when a computer drives, we should make it easy for the computer.
The obsession with putting this junk on the roads with real humans is the single largest downfall of it. It's ridiculous. And then you find that the systems we have made "dumb" generally "just work". Everything from automated train and tram lines, to production factories, to TVs that just play content rather than try to come preloaded with apps and voice recognition and network connections that can get viruses.
I am a *massive* IT guy, I program, I studied computing at university, I was hooked from a young age on these machines. And I would not trust it thinking for itself. If I was to design *any* system it would do what it was told. Sure, that means you have to tell absolutely everything that you want it to do. But I'd rather have an obedient and dumb system than a disobedient "smart" one.
You wanna play with this stuff, do it away from the roads and away from humans and in a controlled environment where the most that can happen is you bump someone's ankle at low speed.
You wanna do something useful... dumb it down to the absolute basics. Like a washing machine with seven thousand programs on it... you just want it to wash the damn clothes. Make the car go from A to B in the simplest way possible where there's no chance of error. That means stop all this automated car junk and - at minimum - designate one lane "automated vehicles only". Stick a crash barrier between it and other people. Jam the cars into it, inches from each other's bumpers. Have them talk to each other, and blindly obey rules about what they do. And label every few hundred yards with a radio transmitter that tells them where they are, what's ahead and what they should do (stop, go, etc.), and the whole system comes to a halt in absence of such instructions.
By the way, it's called a railway.
https://en.wikipedia.org/wiki/British_Rail_Class_139
What I would like the autonomous vehicle people to explain is what actually happens about public transport in the brave new world?
There are a couple of driverless monorail type systems around, but the Class 139 above was fully crewed (driver and guard) when I was on it yesterday.
Seriously, the economics and social aspects of driverless need exploring as well as the regulatory/safety and liability aspects.
Lines on the road don't last too long where I live. Most roads don't have lines at all. Basically you'd be limiting the self driving to a minority of roads. That minority of roads may account for 90-95% of overall mileage, but we are probably not far from cars that handle 90-95% of driving today. That's probably worse than letting people drive all the miles, because if you only drive 5% of the time you drive now you are probably not going to be as good at it - people who are just learning to drive today and never drive more than 5% of their overall miles will NEVER become good at it.
That ignores when the car doesn't know what the hell to do if e.g. blowing snow covers the lines in the road and the car has to come to a quick and unexpected halt because it no longer has its punch card input to tell it what to do.
And nevermind how much more expensive it would be needing to keep the lines painted well enough for the car to see, and the inevitability of miscreants changing the paint lines so they encode something like "detour right" on a bridge.
You're assuming paint.
I'm assuming radio beacon and/or visual display. A laser in a cat's eye, or on a pole on the side of the road. Radio chirps will work through almost any amount of snow, and the car is *in contact* with the road. Never had a Scalextric? Put a great big sprung connector (like.. a train has!) connecting to the road... it'll clear the way and have a direct electrical connection good enough for data and power (even if it blips). Just like... a train does. There are myriad ways to achieve this and computer-vision is the VERY LAST thing you want to lay a human life on. Hell, 4G and a GPS signal on an isolated lane is more than enough. Just stop it mingling with "normal" traffic.
Snow-covered? Continue on using your front and rear sensors at a slower speed until you pick up the signal again. Just like a train does. Or a human when they can't see the road.
Hello Lee,
I love you and want have your babies. And the last time I looked I'm a man. Oh well.
All of this is blindingly obvious if you've had anything to do with software development, in any field, and have been paying attention and been responsible for anythng critical in the real world. I've been there, and the whole AI concept frightens the life out of me.
To quote John Denver: 'It turns me on to think of growing old.' Actually it doesn't, but it's a relief.
Even if it were technically achievable in any of our lifetimes, which I seriously doubt, I also doubt many people actually want it. Think it would help reduce accidents in your country? I suspect proper driver training and law enforcement might be a better use of the time and energy than this horseshit.
Do these self driving systems take into account the poor weather? If there is snow or ice on the ground, does it know this? Will it continue to take a corner at the speed limit, or will it slow down because the road conditions are bad? Do the sensors work in a white out where you can trust that you are staying in lane because the lanes are covered in snow, and yours eyes can't see more than 20ft in front of you.
I guess that comes down to common sense, and I would not let the car take full control in those conditions, but what if some do. They then not only risk their but those in another vehicle.
They might encounter a bit of rain here and there during their testing in CA and AZ, but those places don't have the kind of heavy downpours where you have to pull over due to lack of visibility. They've probably never tested with a single snowflake on the ground.
Musk has a penchant for overhyping features like autopilot and now full self drive. Then it gets delivered in some underwhelming, not-fit-for-purpose form and the stock tanks. Even worse, sometimes people get themselves killed through its inadequacies.
He'd be better off to understate what it can do and only deliver what is achievable. Anyone thinking "full self drive" will produce a car fully capable of driving itself without driver intervention is kidding themselves.
I'm not convinced that any of this is news worthy. Skilled developers with niche and hotly in demand machine learning skills, a Californian tech jobs market that pays silly money, devs jump ship every 2 or 3 years to keep their skills and CVs up to date. A prestige company like Tesla is a good thing to have on your CV. Doesn't mean you ever had any intention to stay there long term. Yawn. Good luck to them.
Airborne software used to be really reliable, but was extremely expensive to create and test and dealt with a much simpler environment. Even so, it used to be subject to massive regulation and certification. Recent events show what happens when that process is 'streamlined' to save money through cheaper dev/test and/or reduced or sub-contracted certification.
What on earth do we think will happen when this sort of (much much more complicated) software gets developed for cars which are relatively cheap, more loosely regulated and developed by companies that focus on saving pennies on their multi-thousand pound products?
In fairness, WayMo looks to be much more conservative than Musk/Tesla in its approach to vehicle autonomy. Musk gets the headlines and kills a few customers. Whereas Waymo just sort of plods along. But they do have record of managing to share the road with normal drivers without undue carnage. It'd be nice to know what their actual plans and expectations are. When do **THEY** think vehicles will reach full autonomy?
Waymo is also conservative in that they drive all their miles in a few small areas, so they can have those areas mapped down to the millimeter. They probably have the car steering around potholes because it already knows where they are, if indeed there are any potholes in the nice neighborhoods they're driving around in.
Those Waymo cars that drive so successfully would have no fricking clue what to do if you took them into a new area with different challenges they've never faced. If you only want a car to drive you around the city limits of Palo Alto and Sunnyvale you'll probably be able to get one before long :)
Including the cliffs they drive over and bridges they drive off when you let not ready for prime time self driving cars drive on roads they've never seen. They have to have safety drivers the first time - plus maybe more times in some cases to try it in the dark, rain, snow etc.
"Recent events show what happens when that process is 'streamlined' to save money through cheaper dev/test and/or reduced or sub-contracted certification."
No. Recent events show what happens when lobbying and flat out corruption allows "self regulation" and regulatory capture.
Boeing has had the FAA in its pocket for quite some time. Look into the 737NG counterfeit parts scandal and the shortcuts taken in the 787 approval process that should never have allowed the lithium ion batteries used to be flown. Corruption is somewhat like a rotting fish and somewhat like dry rot in a house - it starts at the head and works its way downwards - when you see it, it's spread throughout the entire system and it's usually so rotten that the entire shell can collapse with a good hard poke.
That's entirely different to the processes happening with cars - simply because UNLIKE the aviation market which is controlled by a few companies and where aircraft making is a matter of national pride, regulators in every country have a deep interest in cars not killing their citizens.
Autonomous vehicles are coming step by step. Robotic pallets in warehouses have been around for a long time. An autonomous lawnmower is a vehicle when you think about it. ISTR that somebody is currently trialling autonomous shuttle buses on the fenced-off perimeter track of an airfield somewhere. The military are approaching the point where collateral damage from an autonomous fighting vehicle is comparable to that from a GI. The things are eating their way steadily into less-controlled and higher-speed environments. But on the public highway where they could be asked to go anywhere from a primary school car park to a motorway in the rush-hour? If the industry has any sense, low-speed, off-peak delivery vehicles from the likes of Amazon will be sent out to trundle round the neighbourhood at 20 mph or less, then urban taxis at up to 30 mph, all long before hairy supermonsters can be let out willi-nilly. Indeed, if it proves that general intelligence is a necessary ingredient then all bets are off.
Musk may have had some smart ideas once, but he has already picked the low-hanging fruit and the only part of him that is not burning out now is his ego.
"An autonomous lawnmower is a vehicle when you think about it. "
Yeah, but the autonomous lawnmowers I've seen have a wire outlining the edge of the yard, and the mower just moves around inside that area. AFAIK they rely on people having the sense to not step out in front of them, they would gleefully (try to) run me down and chop me up into bits if I stood in front.
"Yeah, but the autonomous lawnmowers I've seen have a wire outlining the edge of the yard, and the mower just moves around inside that area. AFAIK they rely on people having the sense to not step out in front of them, they would gleefully (try to) run me down and chop me up into bits if I stood in front."
That's the point. A wired-up lawn is a pretty dam' controlled environment and despite that the mower is obliged to move real slow. But it's still an autonomous vehicle. However I have been told that if the average smart mower does run into you, it will take the rather less environmentally-friendly approach of backing off and going round you. ;o)
A. I personally don't care if I ever see a self driving car on a public road.
B. Selling this option without insufficient review seems beyond the reasonable limit of disruption/confrontation of the un-evolved rules of 50 years or so ago regarding public safety, DOT approval etc. I will never not drive my own car, and I am damn sure concerned about whatever else is on the road.
C. Real disruption is low maintenance electrics that charge anywhere quickly, recycle completely, and last a very long time, and cost less than the average for whatever market segment is in. Tesla will not accomplish any of this, and doesn't appear to be interested in it. One of the big car makers is going to turn Tesla into the American Alfa Romeo.
ps - I am currently a Tesla shareholder - I was hoping it would recover to enough for me to get out without too much pain. This seems unlikely now...
pps- I don't want any autonomous system developed by the Tech Bro Cabal on the road, ever...
"Real disruption is low maintenance electrics that charge anywhere quickly, recycle completely, and last a very long time, and cost less than the average for whatever market segment is in. Tesla will not accomplish any of this."
Hmm, from what I can see Teslas require almost zero maintenance, no servicing schedule and even the brakes wear a lot slower than an ICE car due to the regen braking.
You can charge from anywhere where here is a charger. There are more public charging locations in the UK than fuel stations and as most people will charge at home overnight the need to actually publically charge will be limited to a few times a year for most people.
The batteries can be at least 50% recycled in a standard plant today but can be used for other purposes without recycling when no longer usable in a vehicle (already being used in a stadium). The batteries are also predicted to last over 400,000 miles so that's generally a very long time.
Tesla also have the advantage of having their own battery factory and supercharging network which make all of this possible and don't rely on various third parties with poor cells to power their vehicles (like the old Leafs that are useless secondhand).
I think most in the industry would agree that Tesla are pretty far ahead in terms of the drivetrain, battery tech, battery production and charging infrastructure. In the next 2 years there should be some decent options from a few manufacturers but there'll be quite some catching up to do and I doubt anyone will try to match Tesla's charging network, they'll just rely on third-party infrastructure.
I appreciate your rebuttle. I didn't make my point well enough.
I am trying to say there is a difference between wedging an electric car concept into the current world view versus adapting an electric car to the current world.
I think Tesla was right to pursue the bulk of the main technologies in the way they did, to encourage adoption and demonstrate viability, and so on.
Tying electric vehicles to autonomous driving is a disservice to the benefits of electric vehicles. I think focusing on viable, inexpensive electric cars for the masses is more important the chasing autonomous vehicle technology.
Imagine what automobiles would have looked like if the oil companies overtly own car companies. Hummer H3 is the subcompact market leader...
As for their market leadership in drivetrain etc. I bet that will be past tense in less than 1 year. Too many competitors in the pipeline, especially at the high end. Measured against the corporate trend to capture the consumer, versus actually produce the best possible singular product, means, in my view, it will fall prey to the kind of constant fall short/big promise delays like MS - the market leader in tech vanity, has demonstrated for decades.
"as most people will charge at home overnight the need to actually publically charge will be limited to a few times a year for most people."
The town planners/construction companies seem to be working against that, I'm afraid. I live in a semi-rural area, and all of the new-build houses being built have no front garden or driveway, with parking areas (i.e. car parks) away from the houses. The local town is Victorian, so entirely terraced with no off-street parking. Neither of those situations allow for home-charging of EVs, unfortunately (yes, you could snake a lead over the pavement over night, assuming you can actually park near your house, but it's not a good idea).
The first example would seem a fairly easy problem to solve. As it is a car park then charging points could be provided in the car park with those spaces rented by EV owners (and expanded as needed).
The terraced house option is more difficult but is already being addresses by others. Some councils are facilitating lamppost charging points, encouraging third parties to provide roadside charging or allowing cables over the pavement with approved cable management to stop tripping. In fact, in the same way that councils allow drop kerbs to be put in, they could allows very thin gully 1cm wide at the surface, but more space underneath, to be requested where a charging cable could be laid in overnight. The local responsible authorities have many available solutions for this if required.
...has lost about 10 per cent of its staff... they have all the needed hardware in place and are simply waiting for the software to catch up...
There is my patented relational blockchain AI database, which Tesla can use right now (right after Tesla bought a license). All Tesla needs to do is to install and start using it. That is, Mr. Musk is right dismissing those who are not able to change themselves and understand this new AI technology, those who continue to write code manually, not seeing that everyday language/ texts can (without their participation) be structured into (in some sense) programs and replaced all their code.
Well, setting unrealistic deadlines for not fully specified complex systems has failed consistently on every software project for decades now, so there's no reason for his magicness to believe that that won't happen because his own sheer brilliance will make miracles happen.
Those with long memories will recall when IBM and Microsoft set aggressive goals for OS/2. Leading to "OS/2 = half an operating system". I tried to use it once back in those days. It was, in fact, truly awful.
IBM did eventually tame OS/2. But only well after the marketplace had moved on.
"So it may turn out that in fact he has been closely engaged in AI for many years."
AI has been many things to many people, but it's mostly dumb, rules-based shit.
And for 99.9% of vehicle operation that's exactly what's needed. Problems usually happen when people DON'T stick to the rules, exceed the limits of the available physics or attempt to put 2 objects in one space - with the worst offenders (as I've noted above) being _slower_ drivers who tend to blame everyone else for crashes but are the cause or catalyst of most of them.
I mean my AI database, which
- uses AI-parsing,
- employs blockchain technology,
- annotates by dictionary definitions,
- deletes lexical noise,
- constructs synonymous clusters.
In short - AI database uses structured texts, which substitute programmms.
People are getting tangled up on science fiction and machine vision problems, and they seem to be missing the forrest for the trees. Not every problem needs to be addressed with LIDAR, GPS and neural nets.
If machine vision systems are having trouble with our lane markings, or lack their of, we could CHANGE them to something that works better. Ditto with signs. Paint is not precious. If a couple of wires or different paint markings could buy several 9s of reliability AND free up the carputers resources to look for more exceptional events, why are we beating ourselves up trying so hard to make a Neural Net figure out out existing and inconsistent markings just from camera inputs?
In most cases we probably wouldn't need to remove the existing signage either, just adding supplemental markings would help.
In many places, there are no road markings - country lanes etc. It just isn't cost effective to mark them.
In many places, there are too many road markings, where drivers get assaulted with too much data, some of it conflicting.
Adding more markings isn't going to be cost effective, or solve confusion, depending on where you are.
Round here, when we complained about the articulated lorries ignoring their restrictions on our unsuitable country roads, the County Council had white lines painted on 2 miles of road. At some points the road surface is only 11 feet wide, but with the line still painted down the middle.
Now there is an edge case for your testing of line following!
This self-driving technology has been very, very successful. It's amazing, in fact.
A large workforce is employed in good jobs and the owner is fabulously rich.
The fact that the product may never work and the Emperor has no Clothes is neither here nor there.
In order for the 'Tesla Network' to function where all the Tesla's go off and become robotaxi's while their temporary owner is at work then FSD is a must.
According to Elon (who must be obeyed) Musk, Tesla are soon going to stop selling cars altogether. You will be able to lease (aka rent) one but once that period is over it goes back to Tesla to become a Robotaxi.
He sees much more money to be made from a taxi service than making cars.
Getting this approved from a legal standpoint will be a huge problem. The issues of liability have not even really been discussed yer.
I am just going to be happy to have a car that navigates controlled-access roads like the interstate, maintaining safe distances and speed, until it notifies me about two miles before that my exit is coming up, then hands back control. I would not trust it on the poorly marked, low-shoulder country roads near my house, or in poor driving conditions like ice or heavy rain. That is my fond wish for my next car, do not know if that is feasible.
"You still have to be able to take over at a moment's notice though, which is presumably what FSD will be for."
Therein lies the problem. FSD does the opposite of what the marketing hype pretends. By having to monitor the performance of the supposedly autonomous system(s) as well as the actual environment, the workload on the operator/passenger is actually increased. Hypnosis By Boredom(TM) will be a contributor that isn't presently being addressed because of the blindness caused by so many shiny new objects.
Forgive my obvious Luditeness but, presumably you want a self-driving car to get from A to B whilst reading, knitting, having breakfast, playing with yourself, etc, right? Why not just take public transport? The end result is the same and it's cheaper than a Tesla. Also, in most cities, cars are the most ineffective and inefficient method of transportation, often beaten by pedestrians, motorcycles, buses, cyclists, skateboards and trains.
I've often thought, whilst walking by stationary vehicles during rush hour, that you may as well join all of these cars physically together to hold, say, about 30-50 people, have them go on predictable routes convenient for the majority and split the cost between passengers. Genius!
You have to schlep yourself to the nearest bus stop, which might be a few miles away in the next village (what, you think everyone lives in a city?). Or you end up forced to travel at a specific time - buses only run once or twice a day to some places. Or you might need to travel somewhere that doesn't have a direct public transport route, and end up needing two or three changes on multiple modes of transport - that can end up getting very expensive (yes, even more than doing the journey in a Tesla) as well as taking a long time. Or you might be carrying goods which would be difficult to take with you on public transport - even a minimal toolbox can be a massive pain to drag around on the bus.
A lot of the cars you see in traffic jams have gone directly from A to B, and for most of the route they've cruised at around the speed limit for the roads, and have hit traffic just for small sections at the beginning and/or end of their journey. Even if there is a convenient bus for that route it may be a faster, less stressful, and more pleasant experience for the person to go by car. It might even be cheaper too, if you only account for the marginal costs of the journey concerned. And if there's more than one of you in the car it's even more of a no-brainer.
Public transport can be absolutely awesome for getting you to the pub and back, for regular commutes for which you've specifically chosen your home/work location so that it's easy, or for occasional trips elsewhere in a well-meshed transport network (e.g most of London). For everything else it sucks binky.
P.S. cycling is even better than the bus. So there.
Human intelligence can never be replicated in a machine, simply because no machines can be built to be like a human. A human driver takes in a lot of information in a glance. Pedestrians, other vehicles and their movements, people seen through windows of other vehicles, people engrossed in their devices, etc. All the little details that trigger our reactions. Many programmers are like teenagers- they learnt to code a few lines and they believe they can do wonders. You need to be intelligent enough to know you’re not intelligent enough.
Waymo and Argo:
At the end of a test day, all the data gets ingested into a data center from the vehicles and the good stuff is analyzed and labeled. Raw data by itself doesn’t have much value for training the machine learning systems that form the core of modern AV systems. The objects in the data that are of interest including pedestrians, cyclists, animals, traffic signals and more. Before any sensor data can be used to train or test an AI system, all of those targets need to be labeled and annotated by hand so that the system can understand what it is “seeing.”
https://www.forbes.com/sites/samabuelsamid/2019/06/19/argo-ai-and-waymo-release-automated-driving-data-sets/
Two major points have been mainly glossed over in the techie rush to debate the minutiae.
(1) If you started today and mandated that all new cars had to be fully autonomous (well, slip that a bit, all new models, what do you mean 737?) it would take a minimum of 20 years to age the current vehicles off the road, probably longer as cars last these days. So given that we aren't seeing any mass market fully autonomous vehicles yet that is 30-50 years for the rollout. Assuming we have ubiquitous technology within 10 years.
(2) There seems to be a desperate push to fit the technology to existing roads. Possibly OK for recently constructed cities where there is a lot of spare land, but not so good for more ancient cities where the roads were designed to accommodate a well driven chariot. To work effectively the road infrastructure needs to be upgraded specifically to support autonomous driving in a phased approach. Once there are enough 'autonomous' lanes on major multi-lane highways and in cities and towns (think bus lanes) then premium buyers probably will pay to make parts of journeys easier. Make the roads fit the technology, not vice versa.
In general these future technology debates, including mandating electric vehicles, are city centric and ignore the rural edge cases. Fine, you have developed an efficient and very light and streamlined vehicle which can stretch its range to 350 miles. Now lets see how far it can go towing a 3.5 tonne trailer/horse box or 2 tonne large caravan. Noting that the answer is not to breed lighter horses (or other general livestock). Given that the towing vehicle has to have enough mass to stop the towed vehicle from overpowering it. OK now let me see that do it all autonomously.
TL;DR the last 10% of having all vehicles fully autonomous (and electric) is likely to take longer than the other 90%.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
