Slashdot Mirror
Uber Expands Driverless-Car Push With Deal For 24,000 Volvos (bloomberg.com)
Uber agreed to buy 24,000 sport utility vehicles from Volvo to form a fleet of driverless autos. According to Bloomberg, "The XC90s, priced from $46,900 at U.S. dealers, will be delivered from 2019 to 2021 in the first commercial purchase by a ride-hailing provider." Uber will add its own sensors and software to permit pilot-less driving. From the report: Uber's order steps up efforts to replace human drivers, the biggest cost in its on-demand taxi service. The autonomous fleet is small compared with the more than 2 million people who drive for Uber but reflects dedication to the company's strategy of developing self-driving cars. "This new agreement puts us on a path toward mass-produced, self-driving vehicles at scale," Jeff Miller, Uber's head of auto alliances, told Bloomberg News. "The more people working on the problem, we'll get there faster and with better, safer, more reliable systems."
I don't see self-driving cars without a "check driver" being legal in any (US) city where rideshare is popular by 2019 (in a year and 2 months). Still a lot of unanswered questions:
(1) How to get them to deal with snow
(2) Cyclists/pedestrians -- none of which behave like normal vehicles obeying traffic laws exactly.
(3) Construction areas
They also currently require very detailed mapping to work -- any truly autonomous vehicle should be able to take info from GPS, a general map of the street network, visual/IR/mmWave sensors, and street markings/signs in order to operate safely. Details of roads change too frequently (by the hour, sometimes) to be reliable.
Not to mention whom to hit in a pinch. The child crossing the street, the ambulance, a lamp post, or the school bus?
The obstacles are legal as well as engineering -- the above is a valid question. Humans can be assigned responsibility as the courts see fit. Autonomous cars will need specific rules of some type.
Also, why Volvo XC-90s? Seems a waste to have a 300 hp guzzle-wagon in an application that will primarily require it to obey traffic laws in a city religiously. Better off with Priuses or even straight electric cars.
They also currently require very detailed mapping to work -- any truly autonomous vehicle should be able to take info from GPS, a general map of the street network, visual/IR/mmWave sensors, and street markings/signs in order to operate safely. Details of roads change too frequently (by the hour, sometimes) to be reliable.
This is one of the problems of GPS maps. None of them can take construction work and the chance of a possible road-closure and/or detour into consideration, much less deal with the fact that there may be no right/left turn in some intersections at different times of the day.
All this information needs to be updated constantly.
READY.
PRINT ""+-0
1: Simple answer is: Don't deploy them in areas with snow until that's sorted.
2: Already taken into consideration on Google, at least. However, you also get to the point where those lemmings just want to jump off a cliff*... A human driver is also likely to kill someone who cuts in front of him/her (physics and all that)
3: Not seeing the problem.
* Yes, I know that lemmings jumping off cliffs thing was faked by Disney. But still...
I think it's great that Uber is investing all this money into creating an autonomous driving fleet but I don't see how this will make money for the company in the long (not to mention the short) term. Along with the $1B+ for the Volvos at list price I think it would be fair to add at least another $1B for sensors and software.
So how does this make sense for a company that $6.6B as of June (https://venturebeat.com/2017/08/23/uber-is-still-burning-cash-at-a-rate-of-2-billion-a-year/) and is burning cash at a rate of $2B/year *before* doing this?
Mimetics Inc. Twitter
Negotiation between human drivers, cyclists, other drivers, and pedestrians is complex and non-verbal.
I suspect this isn't being paid for right now -- more of an option to buy at a specific price/quantity over time.
This being said, when the economy goes sour (not if, but when -- we're overdue for a recession), a lot of tech firms will burn through their cash, crash, and burn.
Jokes about Volvo drivers
READY.
PRINT ""+-0
Who rapes the female passengers now? Perhaps a case for teledildonics at last?
In the first case, you brake, how many times have you had to make that decision in traffic and what would you do? In the few cases I've been in life-threatening situations in car, I try to brake first and then aim for the lightest/smallest object or anywhere that has more open space - you don't really get to think about the repercussions of a moral decision when your car is doing 360's on a frozen highway.
Most likely because Volvo is the only manufacturer that seems to actually working on useful sensors in cars (not just overpriced overhyped TensorFlow chips).
Custom electronics and digital signage for your business: www.evcircuits.com
Uber's market cap is about 50 billion. A 2% move is a cool billion.
That said: If they aren't completely autonomous and require an 'emergency human', they will move the stock a lot more than 2%, but not in the direction they want.
If I was Uber, I'd try and get Volvo to accept stock for the mall utility vehicles, or carry the note in Volvo financial (don't know if it exists).
John McAfee 'It was like that time I hired that Bangkok prostitute; to do my taxes, while I fucked my accountant'
My GPS system does get live traffic updates including road closures and detours. There's this thing called cell phone modems, they allow you to exchange data on mobile devices.
Custom electronics and digital signage for your business: www.evcircuits.com
>>Not to mention whom to hit in a pinch. The child crossing the street, the ambulance, a lamp post, or the school bus?>The obstacles are legal as well as engineering -- the above is a valid question. Humans can be assigned responsibility as the courts see fit. Autonomous cars will need specific rules of some type.
This is true.
This being said: Firstly, the position is that the company that makes the car is the driver. (That's Google's position, anyway)
Secondly: Unless we're talking about civil litigation (go to the company that owns it) what's the point in criminal charges? Criminal charges are there to punish someone who should have known better. Someone who was negligent or intentionally damaging.
Whatever happens, happens, and the courts ascribe liability -- to the driver. The little guy, generally with shallow pockets.
The equation changes when large companies with deep pockets need to make that decision ahead of time, and they (or their insurers) take over liability.
>> This is one of the problems of GPS maps. None of them can take construction work and the chance of a possible road-closure and/or detour into consideration> much less deal with the fact that there may be no right/left turn in some intersections at different times of the day.>All this information needs to be updated constantly.
This last is very true. However, shouldn't be onerous. If you can put a sign in (and put a work order for that sign in), you can update a database.
Also, object recognition is actually really good now. It's not perfect, but for something that's both consistent and easily recognisable like a sign. Fantastic.
(1) How to get them to deal with snow
Why does every self-driving car denier tout this out as "IT'LL NEVER HAPPEN" evidence?
Humans can drive on snow with 2 EM sensors that are limited to the 'visible spectrum' and an IMU that gets messed up rather easily? Especially when on snow?
With the range of sensors and their sampling rate I expect them to be much, much better on snow.
Quick, count the last times you have ever run into a child crossing the street while an ambulance was bearing down on you with a school bus?
Whom to hit in a pinch? A celebrity, an asteroid, a school bus full of nuns or a tumble weed in downtown NYC?
No -- it gets updates as they're updated, not as they happen.
Also, how detailed will the info the cars hold be? Example -- I grew up in a town that flooded. One underpass had a dip in the road that could hold 3 to 4 feet of water, more than enough to stop a car and possibly harm the occupants.
Would an autonomous car stop at any amount of water detected on a road? (Not necessarily correct.) Would it try to drive through? (Not correct either.) Will it know the depth of the bridge and the flooding behavior?
Driving in anything but perfect conditions isn't simple.
Put in an LCD that's capable of flipping someone off. That, plus indicators, covers close to 100% of communication requirements. More seriously: A cyclist and pedestrian don't want to step in front of something that weighs a ton that's travelling at high speed. I don't care who's driving, it doesn't make a difference. Inertia says that the only reason that person doesn't become a Darwin awards entrant is because the method of death is so...... Pedestrian....
No, but I've been in the following situation.
(1) hit the deer, possibly damaging my car and ending up with a buck through the windshield
(2) cross the center line on a hill with poor visibility.
As a pedestrian, I think I'll carry an inflatable balloon that looks exactly like a block of concrete to the robocar's sensors. Pssssssssssht goes the CO2 cartridge. SCREEEEEEEE, SCREEE, CRUNCH (crosses street).
So what did you do?
Because as a human you're slow, you think you only had those 2 options.
Hard brake and swerve to the right onto the shoulder and only take out a front quarter panel.
Hard brake and avoid the hit completely since it didn't waste an extra half a second to decide. Brake 3 wheels and accelerate one and have the car whip around the buck. Feather brake & accelerator to time no collision.
>>No -- it gets updates as they're updated, not as they happen.>Also, how detailed will the info the cars hold be? Example -- I grew up in a town that flooded. One underpass had a dip in the road that could hold 3 to 4 feet of water, more than enough to stop a car and possibly harm the occupants.>Would an autonomous car stop at any amount of water detected on a road? (Not necessarily correct.) Would it try to drive through? (Not correct either.) Will it know the depth of the bridge and the flooding behavior? Would a human who doesn't know the area? This being said: One single car would be caught out by an area. That would be logged and never be a problem for anyone again.
No shoulder on that road. Would crossing the centerline (putting other traffic/cyclists/pedestrians at risk) be acceptable if you were an autonomous car?
Driving in anything but perfect conditions isn't simple.
Running a steam engine in anything but perfect conditions isn't simple: 1600s.
Running an internal combustion engine in anything but perfect conditions isn't simple: Late 1800s.
Flying an airplane in anything but perfect conditions isn't simple: Early 1900s.
Driving a horseless carriage in anything but perfect conditions isn't simple: Early 1900s.
Communicating over wireless in anything but perfect conditions isn't simple: Early 1900s.
Running a computer in anything but perfect conditions isn't simple: Mid 1900s.
I wonder where we would be technology wise if everyone listened to your type.
I can't speak for deer, but the official line for kangaroos is: Hit the thing. You're far less likely to be killed.... Don't swerve. Ever.
1. What did you chose to do?
2. Why would an autonomous car make a worse decision than a human?
3. It's a strawman because you made the decision based on very limited data and sample rate. It's a scenario that the car would have avoided completely. The car would have seen the deer on FLIR or LIDAR. The car would have had more reaction time to stop completely.
Predictions generally outpace reality, though.
In the 60s, we were supposed to be living on the Moon and going to Mars by the 1990s. Self-guiding (maybe flying!) cars were 10 years away in 1965. In the 70s, anyone would be able to buy a ticket on a supersonic jet by 1985. We were supposed to get our (snail) mail by missile...
It wouldn't make a worse decision than a human. But the "driver" will have deeper pockets and more to lose, so there will need to be a legal structure as to what to do in such a case (and others).
If your exit strategy is an IPO, then that strategy makes perfect sense.
The grander the promise, the more money you'll need for your IPO.
Yeah.... Lots of places don't get snow. I'm a Queenslander (Australia) I didn't see snow until I was 20. I've also lived in England..... Humans are pretty bad at dealing with snow as well. I mean, they're just terrible.
This -- milk it while you can.
.... Ok, police should watch you. You clearly have sociopathic tendencies :P
More seriously.
Doing that stuff will work on a human driver as well.
If you think our brains will have time to think "No, that's not a REAL object in front of me, I'll just go through it" you're dreaming. In those timeframes, we act on instinct.
When discussing this stuff, so many people seem to have the idea that humans are capable of making complex decisions and seeing through trickery in timeframes of half a second.... Or even a few seconds.
We're just not.
For a computer, on the other hand, those timeframes are an eternity. Physics is the bigger problem.
Tax benefits and possessively some type of exclusive agreement with Volvo so that other companies can't buy the same thing.
In the first case, you brake, how many times have you had to make that decision in traffic and what would you do? In the few cases I've been in life-threatening situations in car, I try to brake first and then aim for the lightest/smallest object or anywhere that has more open space - you don't really get to think about the repercussions of a moral decision when your car is doing 360's on a frozen highway.
Of course you brake. But that's occasionally not enough. And the engineers DO get time to think about the repercussions while designing the AI to drive the car.
Most likely because Volvo is the only manufacturer that seems to actually working on useful sensors in cars (not just overpriced overhyped TensorFlow chips).
Did you read the summary? Uber will be adding their own sensors.
bloomberg called them a "taxi service".
One of the problems...
Do we really want to create "efficient" traffic that's easy for computers to navigate, at the expense of usability for pedestrians and cyclists?
https://www.youtube.com/watch?...
Try crossing the street or cycling in this mess? Will we all be bound to our cars just to walk across the street?
Hard brake and swerve will likely lead to spinning out and having a massive crash. Especially since shoulders rarely have the same grip as the main road.
Not to mention whom to hit in a pinch. The child crossing the street, the ambulance, a lamp post, or the school bus?
How about you hit no one. The areas they've deployed to already limit their speed at 25 MPH. If something goes wrong at 25 MPH, you stop. That's not rocket science.
Plus, you speak like it knows the difference between a child, a raccoon, or a lamp post. Computer vision (disconnected from the cloud) is nowhere near that advanced yet. For a car to drive itself, it needs to know where the road is and if there is an obstacle. That's basically it
What you're speaking about may come into play 10 or 20 years from now, but by then, many problems will have been figured out already by just analyzing accidents that occurred in the past.
Humans are extremely good in dealing with snow -- think about it: A slow CPU with four main sensors is currently doing better at a task than many faster CPUs with a dozen or more sensors.
>>It wouldn't make a worse decision than a human. But the "driver" will have deeper pockets and more to lose, so there will need to be a legal structure as to what to do in such a case (and others). THat's what insurance is for (even for large companies like that... Although they might self insure) Also, road injuries and fatalities are expected to plummet... Which should make insurance far more affordable.
Volvo needs to deliver and expect there to be performance requirements which if late or below spec can cancel or get a good discount. The advance order helps Volvo allocate resources to new tech and potential product. Seems reasonable gamble. Can still use them as semi auto to begin as issues resolved, a safer ride.
2. Why would an autonomous car make a worse decision than a human?
Because sensors still aren't as good as human eyes at recognising many hazards. Also, an autonomous car can't think about the best option for the situation unless the engineers building it anticipated that specific situation. They're never going to be able to anticipate everything.
What happens if there's a rock slide and there's a boulder rolling down the hill towards the road, potentially colliding with the car? Were the sensors designed to detect that? Was the AI designed to predict boulder trajectories on uneven terrain? Is stopping in the potential path of a rock slide even a good idea?
3. It's a strawman because you made the decision based on very limited data and sample rate. It's a scenario that the car would have avoided completely. The car would have seen the deer on FLIR or LIDAR. The car would have had more reaction time to stop completely.
Would it have known it was a deer? LIDAR doesn't distinguish types of obstacles. Maybe it was a bird. Maybe suddenly stopping with superhuman reflexes while there's a big truck following you isn't such a good idea.
I need to be very clear here: And cyclists are a gray area. However, they should always be conscious of what's going on around them, and not do anything unpredictable. If you're calling it "Traffic" pedestrians SHOULD NOT BE IN IT. If you've got pedestrians wandering through traffic, there's a bigger problem.
This being said: One single car would be caught out by an area. That would be logged and never be a problem for anyone again.
Assuming the car survived long enough for the data to be uploaded. Also, what if there's no mobile signal?
Also... That mess is the creation of human drivers. How is replacing them with autonomous drivers going to make things worse, exactly?
See the video of NYC traffic... https://www.youtube.com/watch?...
Sorry, jump to about 1:05 in the video for the "mess" I describe -- autonomous cars flowing continuously around each other without lights or stops.
>>Because sensors still aren't as good as human eyes at recognising many hazards. Also, an autonomous car can't think about the best option for the situation unless the engineers building it anticipated that specific situation. They're never going to be able to anticipate everything.>What happens if there's a rock slide and there's a boulder rolling down the hill towards the road, potentially colliding with the car? Were the sensors designed to detect that? Was the AI designed to predict boulder trajectories on uneven terrain? Is stopping in the potential path of a rock slide even a good idea? When I'm driving in a situation like that, I'm looking ahead of me. The rock is coming from my side. Seriously, where are you driving that this is a major problem though? Is boulders going across your path at speed a normal occurrence in your daily commute? There seems to be this absurd obsession with coming up with unusual scenarios to prove that AI driving is useless when human drivers are causing over a million fatalities per year (As of 2010, anyway. I'm guessing by now it's around two million, but whatever. This ignores accidents as a whole. This is just failures that result in a death) It doesn't need to be perfect, it needs to be better, overall, than the meatbags currently driving. Which is not nearly as difficult as making sure it's perfect in scenarios in which human drivers are likely to fail as well.
The technology is such these days that all the examples you site would be probably better detected by computers than they would by human beings. Driving a car safely is actually a very deterministic process, it a simple process of data collection and applying very simple rules e.g... While approach child on side of road, decrease speed by 10% ...
How many times have you seen some idiot in a car get in the way of an ambulance or driving through an intersection and nearly colliding with an ambulance? Well the day will come when driverless cars know the position of every emergency vehicle and react to make way for these vehicles before they would even be within sight of a human.
I can see the day coming when they start to think about banning humans from driving vehicles as they will be regarded as a hazard compared to automated vehicles. I don't think people who make these kinds of arguments against driverless vehicles have much of an understanding about technology or software development.
And possibly quantum effects at the neurological level. Too lazy to look it up, just do a search. What this might mean is the current crop of AI is biomimicry based on a flawed or incomplete premise. So, no, the quality of self driving cars will asymptote until itâ(TM)s not quite good enough, companies will get tired of pumping billions into a black hole and we can all wonder where the self driving cars are in 20 years.
And computers are already doing exactly this. Such information is already available in google maps with real time data collection they know where the congestion is and the traffic hazards are.
> (2) Cyclists/pedestrians
Much less militant pedestrians and cyclists like we have here in Seattle. I drive work vehicles sometimes in downtown Seattle, and even being as slow and careful as I can, I still almost hit them. Several times, I've had pedestrians walk into my car while I was stopped or cyclists that swerve into my work truck when I'm going straight in a lane. I don't see how an autonomous car can do better.
Also, there's the issue of too many signs and traffic signals. There's one intersection near my office that has more than fifty signs and lights. Even though I drive through it regularly, it's still scary.
What ever happens, Uber go BOOM. Smells like running up cash flows and propaganda whilst planning to financially bail out. "We don't knwo what happened and all of a sudden it went bankrupt, yep, uh huh".
Chaos - everything, everywhere, everywhen
>>Assuming the car survived long enough for the data to be uploaded. Also, what if there's no mobile signal? So, basically, you're assuming that this dip would result in immediate and catastrophic destruction of the car and absolutely everything inside it. Does this dip make the car spin off into a constantly running compactor? That's about the only scenario I can imagine in which that sort of total destruction is likely... Well... That or an active volcano... As far as no mobile signal goes... Those places are pretty rare these days. Even then, it seems a safe bet that if an accident that catastrophic happened, someone would fix the root cause in the road's flawed design. You're really stretching to think of scenarios in which this is going to do a bad job.
Humans, as a whole, are terrible on snow. Their knowledge isn't collectively cumulative. Every year you have to train a new fleet of drivers at 16. What is the rate of
On the other hand autonomous mini rally cars have taught themselves to power slide. With a software update every vehicle going forward could all know how to do that.
Generally, this is a very computationally intensive approach, but AutoRally can calculate an optimized trajectory from the weighted average of 2,560 different trajectory possibilities, all simulated in parallel on to the monster onboard GPU. Each of these trajectories represents the oncoming 2.5 seconds of vehicle motion, and AutoRally recomputes this entire optimization process 60 times every second.
That's better than any human can currently do and will only continue to get better.
Seriously reading through the majority of the comments that list inane things like "what about children crossing the road?", "what about cats and dogs?", "what if there's an ambulance?", "what if the road is blocked and there's construction work, how does GPS work then?"
I mean seriously do you guys have no understanding of information systems?
To those of you think driverless cars are too hard, and they can't possibly work, just watch, the only issues driverless cars will have will be trust, in the same way people trusted a horse and cart over a car 100 years ago.... The technology is already all here and those who understand it know that driverless cars will be safer and result in less congestion than roads filled with cars driven by people.
Welcome to the 21st century.
The number of traffic jams and accidents I saw makes me treat that statement with extreme suspicion. Also: Lots of places don't have snow. Huge amounts of them. Even those that do only have snow at particular times of the year (Y'know... Winter) - The exceptions obviously being the arctic and antarctic. Neither of which are known for their heavy traffic. Key word there is also "Currently" - This has a historical tendency to change.
You're right. We have reached the pinnacle of human technological achievement. Pack it up. New stuff takes too long.
But I repeat myself: I wonder where we would be technology wise if everyone listened to your type.
1 billion dollars will look like very cheap compared to the value of the market they'll have when the only way for an average joe to get from A to B is via Uber .
I don't understand why you were voted down. Heading north on 4th is a nightmare where the city grid changes orientation. You have to pass over both Stewart and Olive in the same intersection a block west of the start of the monorail. There's a lot of traffic east on Olive since that's how you get to Howell to get on I-5 south. I drive through that intersection several times a month for just over a decade, but it's still confusing.
Huge, society-changing ones that require massive changes do. We tend to underestimate the smaller ones. Ask anyone who's been put out of work by a robot.
All it needs now is a chat bot to argue politics/sports/religion/weather with the passenger... And how's it gonna hook you up for a quick bump,eh?
>>Sorry, jump to about 1:05 in the video for the "mess" I describe -- autonomous cars flowing continuously around each other without lights or stops. Do you commonly walk into the middle of a busy intersection? Is this a normal thing in your daily walk? If so, how are you still alive? Personally, in a place like that, I like to use this thing I call "Pedestrian crossings" Where there's this super convenient button that makes a light change colour. Traffic then stops. Sure, autonomous cars won't need them for dealing with each other, but they're still going to be around for humans who don't share your deathwish.
For a start, their reaction time is a tiny fraction of yours. And their ability to maintain focus on multiple things puts yours in the dust. For a second... That's not really a problem with the cars, but idiot pedestrians. I mean, your car was stopped. It's pretty clear that they've simply got a death wish. What will be proven, however, is that it wasn't the car's fault, but that the idiot walked in front of a ton of metal going at speed. Doesn't matter who the driver is in that case, but at least there's no ambiguity about whose fault it is.
>>Would it have known it was a deer? LIDAR doesn't distinguish types of obstacles. Maybe it was a bird. Maybe suddenly stopping with superhuman reflexes while there's a big truck following you isn't such a good idea It would have known how big it was and it was there. Which is about all that we're capable of processing in a split second. If you're going to try and claim that, while doing this, we'd also be conscious of the truck behind us and acting accordingly, I'm just going to say you're off the planet. Although I would take this as an opportunity to point out that that's a good argument for automating the truck as well. So both have superhuman reflexes. Seriously, when you say something, try and think real hard about what an actual normal person is going to do in that same situation.
Normal? As if anything about NYC traffic is normal. https://www.youtube.com/watch?...
Normal? As if anything about NYC traffic is normal. https://www.youtube.com/watch?...
So human drivers have failed. I dunno, dude. It's getting pretty tiresome to respond to your (and other's) random specific scenarios, so many of which humans already fail hardcore in. Or to your weird ones that rely on stuff like it currently being OK for pedestrians to just walk out into busy intersections.... It's not. If NYC is difficult, it doesn't get done yet. It's got about a 10th of a percent of the world's population. Do it in easier places, learn and expand. If you come up with a place where it's unsuitable, ask yourself the percent of the world where that isn't a problem. It starts in those places first and we learn. This won't be perfect. We're not perfect either. It only needs to be better than us. Over all better than us. Not better than us in every single specific scenario you can think of, but globally, lead to a reduction in fatalities.
"Most of the world" is actually closer to NYC than L.A. in pedestrian/car/cyclist dynamics. See traffic in most of Asia, India, or Latin America.
The difficulty in dealing with winter roads is that the conditions are not uniform. It may be dry concrete for the most part, then half a block of ice, followed by a light layer of snow with ice underneath.
Humans can see the road up ahead and identify a lot of this, adjusting their speed before it comes up. If the computer can do this as well, then it has a chance of handling things properly. If it can only see obstacles ahead, it's going to have a tough time with the changing roads. Can they detect potholes yet? Some of them are wheel destroyers.
"Most of the world" is actually closer to NYC than L.A. in pedestrian/car/cyclist dynamics. See traffic in most of Asia, India, or Latin America.
Why do you think this is going to start in poor countries? You pick the low hanging fruit first and expand as you learn and technology improves. Why does this need to be explained?
So there's a bigger problem. Or, as I've said previously. You do NYC later. Heaps of cities around the world this can be started in.
In 1972 the Lockheed L10-11 flew from Dallas to Palmdale (including takeoff and landing) with no pilot. I still don't see an airline without pilots. If anything I'd think trains would run without a pilot and yet they have them too, and they run on fixed tracks. I think self-drive cars may be further out than anyone expects.
It wasn't a literal problem for you to solve, you fucking imbecile. It was an illustrative point, and only dipshits like you can't comprehend what was asked.
Compare the relative value of a pilot's salary with the contents of a plane and a taxi driver's salary with the contents of his taxi.
As an Uber driver, what's it like driving a nail in your own coffin?
Well, the response to that point is: If a human driver has time to consider that, then there probably isn't going to be an accident anyway. Accidents are normally a surprise. We've got a split second to react. If we're still weighing the moral pros and cons, we've probably already hit whatever was directly in front of us/
If your car is doing 360s your licence should be revoked. Immediately. There is absolutely no reason for that to ever occur. Ever.
In the first case, you brake ...
... and the SDC will apply the brakes 1500ms sooner, which will make a huge difference: 130 feet at 60 mph. In these sort of "panic" situations, the SDC will almost always outperform a human. Reaction time is more important than contemplative analysis.
You _should_ push your car past its limits in a wide open empty paved space. That often ends in a spin, give yourself room. If you don't know the limits, you likely trip over them or never get anywhere near them. First step is finding them, then you can push them.
360 plus generally requires a little power to keep the drive wheels spinning. Fun though, might have to turn off traction control.
Your right, you should never let the cops catch you doing donuts.
John McAfee 'It was like that time I hired that Bangkok prostitute; to do my taxes, while I fucked my accountant'
Why do you think this is going to start in poor countries?
Looser regulations, and far fewer legal expenses.
The line is elk/horse/cow (depending on how high your vehicle is), elk and larger are likely to bounce once on the hood and then go through the windshield, smegging the occupants hard.
You still don't cross into traffic, but any daylight is an escape route, even if it puts you in the woods.
Big deer can total the car, but you can get another.
John McAfee 'It was like that time I hired that Bangkok prostitute; to do my taxes, while I fucked my accountant'
Is a paper bag in the road an obstacle?
What about one with half a cinder block hidden in it? (Kids are assholes)
On the surface analysis, clutter that moves, leaves and such? Drive over it or grind to a halt in fall. Same for blowing snow.
Kids moving toy? OK build an AI that identifies kids toys in an image. Humans will infer the kid chasing the toy. Think about doing that with AI.
John McAfee 'It was like that time I hired that Bangkok prostitute; to do my taxes, while I fucked my accountant'
Yeah you’d think electric cars although Volvo does have a hybrid version. I’m guessing for it’s safety ratings.
Creeper, _crawling_ past the group of 200 kids on the playground (0.9^200=7E-10). Going to get arrested, weirdo.
John McAfee 'It was like that time I hired that Bangkok prostitute; to do my taxes, while I fucked my accountant'
Reasonable. Can't say that hitting any of those is a common problem here... Kangaroo on the other hand... However, the centre is a bit lower. They're still solid masses of lean meat, bone and muscle though.
Hasn't been my experience when living in the midwestern tundra. 'They' forget how to drive in snow and apparently have to relearn, every goddamn year. It's kind of amazing.
In CA you kind of see it every year with the start of rainy season.
And don't even talk about someplace in the south that gets it's once every 20 years dusting. Just fuck!
John McAfee 'It was like that time I hired that Bangkok prostitute; to do my taxes, while I fucked my accountant'
Looser regulations, and far fewer legal expenses.
Lower ROI as well. A thing something someone calling himself Shanghai bill should be aware of is that in poor countries, taxi drivers are cheap as chips. Which means there's fewer saved dollars (yuan, whatever) from replacing a driver. A thing someone calling himself that should also recognise is that a predictable legal system is far more valuable than lower legal expenses. Finally, a thing someone calling himself that should understand is that for China in particular, it's cheaper to kill the person than to simply disable that person.
If you can see it, there's a very good chance the computer can see it. Also: The people designing this are probably quite aware of its limitations. We're at the dawn of this technology. It'll start where it's easy and expand to where it's hard.
Lower ROI as well. A thing something someone calling himself Shanghai bill should be aware of is that in poor countries, taxi drivers are cheap as chips.
Yep. I have family in Dominican Republic. In the US, they have an electric gate when you leave parking lots. In DR, I've seen one guy taking payments, the other lifting the gate. By hand.
Gas is more expensive than in the US, but labor makes up for it. Cost of a 2.5 hr taxi ride last time I was there was less than a train ride of the same duration can cost in the US.
When labor is cheap, the urge to automation is much less.
Another example -- will A.I. know to increase following distance (dramatically) if something is about to fall off the vehicle ahead of it? Or if the vehicle ahead is a truck whose tire is starting to smoke and is about to throw chunks.
>>Looser regulations, and far fewer legal expenses.
>Lower ROI as well
spotted the genX American who never travels except to Mexico. yeah all south america is poor peoples. kek.
>Lots of places dont even have snow
Sure thing princess. Tell us about your travels through Canada, Europe, Russia, and Asia..
It's not that it'll never happen, it's that it won't happen within the week like all the crazy people saying we have fully capable AI-powered cars right now and the only thing stopping them is legal obstacles. The latest crop of cars which vibrate your steering wheel when you're near the side of the road can't properly detect the lines in snowy or slushy conditions. They can't handle any roads which are dirty with anything and not a single company has put out a video showing their car capable of handling it, so we should assume they can't as such as video would win them tons of positive PR. Knowing where the sides of the road are is an important aspect of driving.
It honestly should, realistically it should only charge costs for fuel with a small % profit. I don't see that happening though in this capitalistic landscape. They may start charging less for the self-driving cars but once there are no more human drivers I see them pushing the price up to cover some imaginary cost. Automation is going to kill lots of jobs, I have no problem with this as it's just the way things go. I do have a problem with it not changing the prices of products and services. Honestly does anybody expect automation to push prices down as it realistically should?
Rich cunts aren't in the habit of handing shit out for what they are worth.
Is a paper bag in the road an obstacle?
They're not using visible light.I understand that LIDAR can see through hedges. So I'd guess if it returns empty on the appropriate wavelength, it's not a problem.
What about one with half a cinder block hidden in it? (Kids are assholes)
Isn't this something that's a problem for a human driver anyway?
Kids moving toy? OK build an AI that identifies kids toys in an image. Humans will infer the kid chasing the toy. Think about doing that with AI.
I'd imagine all it needs to know is that something is moving in front of it. Also, I've got a kid. Even with today's human drivers, I don't let my kid (a four year old) play near the road. Why? Because I know there's still a very good chance my daughter will run out in front of that driver and get blatted. I'm really not seeing how this is worse
This is not true. Last night there was construction. I was going to take my usual route, but my satnav advised it was closed.
At first? Probably not. Eventually, if that actually becomes a significant problem. Probably. Does a human have a good track record for recognising either? (Personally, I've never actually seen a truck whose tire is about to blow. If I saw one tomorrow, I probably would't know what I was looking at, and consequently wouldn't realise I should back off) I'm also not sure I've ever had something gradually loosen and start to fall off a car in front of me. With high wind and all that, when it's happened, it's happened pretty quickly. Still clutching at straws trying to prove this thing isn't perfect? Well... It isn't perfect. But it doesn't need to be to be worthwhile.
When labor is cheap, the urge to automation is much less.
Which goes back to making me wonder why you bothered mentioning those countries. It won't be tested there. It'll be tested in ideal Western cities (there are many) where the technology will be refined, will become cheaper, and from where it'll expand.
I used to peripherally work in the area a decade ago, and standards and methods for communication were being established to gather congestion data from mobile phone 'pings' and traffic systems, primarily UTMC. We were looking at the use of what we called neural networks, but is now deep learning, to recognise patterns in traffic networks, to allow actions to be taken, in this case at UTMC level.
Sure thing princess. Tell us about your travels through Canada, Europe, Russia, and Asia..
Europe and Asia. Quite extensive. Normally there's no snow. In Southern England, whenever there was snow, everything died. Places like Singapore and Hong Kong are also fairly famous for not having snow.
Even then, why... Just.... Why.... do I need to point out over and over again that the difficult places won't be the first ones these will be done in? Low hanging fruit will be picked first, technology will be refined, problems will be overcome, and as it progresses, it'll go to the difficult country.
Just... Why do I need to point this out over and over again? What part of it isn't so obvious that you feel the need to ask the question?
I can see why you're posting anonymously.... I would, too, if I was being that daft.
On average, yeah, it is.
The top crust are rarely taxi drivers.
Also, spoken like an American with an American's knowledge of geography. Asia and India are rarely regarded part of South America
Moron.
Assume a driver costs $40k per year.
That's $960,000,000 per year for 24k drivers.
Which means that this would pay for itself in a little over a year. Ok, this is a bit simplistic, but I doubt the real number is far off... Almost certainly not more than a two year break even point. With this point declining for the next batch.
No, it's not suitable for everywhere.
However, that doesn't change the fact that there are a lot of places where it is suitable. Starting off in cities in tropical or subtropical areas. Expanding from there as technology improves.
I'm not sure about where you are, but in my city (Brisbane) I can't think of many roads that aren't in the middle of nowhere that are so covered in filth the lines are invisible.
oh man so many problems with this. i drove uber "part time" 20-40 hours a week for 3 years, and with over 2500 rides, i have seen some shit. Like the drunk guy who got in, did not put in a home address, mumbled something that sounded like the name of a neighborhood, and passed out. when i got to what i thought was his neighborhood, woke him up for directions, how is an AI car going to do that. Had 3 people vomit in back seat, how do you monitor that? several times a group would get in, argue over where to go for dinner, and wound our way to what seemed like 3rd or 4th possible destination. or the times the passenger is 50-100 ft down the wrong way on a 1 way street, trying to get them to walk to the car could be comical, when they have absolutely no idea where they are or sense of direction. Will this new system have something to guide the cars on where to go when have downtime, to try to be in best spot for a pickup, or will system strive for a coverage map. or just knowing the city, and which way is best, even when gps says go another way. so much human element trying to get people places when they have no idea where to go. will need at a minimum live voice chat to a human to help running 24/7.
So, how many of you still think that Uber is about providing good opportunities for people to earn extra money driving for Uber?
Uber is about making as much money as possible for a few people (the executives and the other shareholders) and nothing else.
All those people who have given up some other line of paid work to drive for Uber are going to be doomed.
Think about that before taking out that multi-year car lease that you'll pay for by driving for Uber.
"For a successful technology, reality must take precedence over public relations, for Nature cannot be fooled"
Yes, it gets updated. By the traffic signs already in use that get put up or removed. The new rules are in effect immediatly.
So your self driving car has to be able to read and understand the traffic signs. If it can't do that, it's not ready to be used on the road.
Self driving cars have to be able to deal with the roads and signage as they are today. No excuses.
They're guessing their tactic of "it's easier to get forgiveness than permission" extends to dumping self-driving cars on the road and hoping no-one has the gumption to tell them to do one. I'd say it's a Hail Mary but it's all they've got at this stage.
I had a dream, bright and carefree, but now there's doubt and gravity
Seen the Will Smith irobot movie?
He was rescued by a robot while a child was allowed to drowned, cos he had a slightly higher % chance of survival.
So even if you have the full maths on who to rescue, in terms of logic, is that always the result you will want?
It's a question that can never have a real answer.
Er, it's not just the hardware. Its *never* just the hardware. It's the intelligence in the software that makes or breaks a system. The software isn't there yet.
Will a driverless car take instructions from a cop in the middle of the road, directing traffic in an emergency?
Add, not replace. Volvo's platform is likely well suited for this.
If your reaction time is more than 1,500 milliseconds, please never get behind the wheel of a car.
All I want is a secure system where it's easy to do anything I want. Is that too much to ask ~~ Randall Munroe
If something goes wrong in an automated car e.g. sensor failure, the software has the option to pull over and stop, or, in extreme circumstances, just stop. Not an option on an aeroplane.
All I want is a secure system where it's easy to do anything I want. Is that too much to ask ~~ Randall Munroe
I started to write about matching driver-less cars to a riders inputted source and destination ... Then thought about being able to ask the driver-less car to get out of the car when it just stopped - note, no talk about crashing , as I'm confident that the driver-less Uber can 'just' stop' - because of all the random stuff that just happens every day. Then considered my major frustration just trying to get Alexa on my phone to understand a simple ask to play an album by the group with the same name ... And thought : if I had lots of time, I'd definitely click the 'ok to send me a driver-less Volvo' box. If I needed to get somewhere reliably - which is almost always the case , I'll let another set of riders experiment.
Google's vehicles can already deal with cyclists, pedestrians and construction areas. In fact they do it better than most humans.
They are now starting to test in poor weather, including snow. They actually have some advantages over human drivers in snow, for example because lidar can see the wrinkle where the curb is buried better than human eyes.
const int one = 65536; (Silvermoon, Texture.cs)
SJW, n: "Someone I don't like, and by the way I'm a fuckwit" - AC
Yes, there is a place where walking around and through traffic , not at cross walks , and against signals , happens millions of times a day. It's called NYC - I'm one of the millions - and probably happens in most other major US cities ...
Assume a driver costs $40k per year. That's $960,000,000 per year for 24k drivers. Which means that this would pay for itself in a little over a year. Ok, this is a bit simplistic, but I doubt the real number is far off... Almost certainly not more than a two year break even point. With this point declining for the next batch.
And it's not just the monetary cost of paying the drivers. It's also the hassle of dealing with HR. Fewer (or no) drivers equals fewer headaches with unionization drives, sexual harassment of passengers, etc. etc.
Actually, your second question answers the first one :
Not to mention whom to hit in a pinch. The child crossing the street, the ambulance, a lamp post, or the school bus?
Also, why Volvo XC-90s? {...} Better off with Priuses or even straight electric cars.
That's why the Volvo (as opposed to one of the current popular electric car brands).
The current fleet of Volvos actually on the streets (which are already street-legal, in production, and driving around your city) have among the best forward collision avoidance systems (FCAS) currently on the market, and have been for quite a few years.
In case of child/ambulance/lamp/post/school bus, FCAS will slam the break in time and try to stop before hitting the obstacle, or at least to decelerate as much as physically possible, to hit it with the lowest velocity if a complete stop cannot happen within the remaining time/distance to collision.
(That is, unless the driver actively tries to override the vehicle's decision by flooring the accelerator. But even if the driver does something stupid like this, the car still sounds a loud alarm and flashes alert light on the wind shield)
Volvos have, year after year, had the best track record of such security features. They even still slightly out-class what's available currently from Tesla (which has a few situation where the car didn't reliably detect a truck trailer, or didn't take the decision to stop before hitting a curb). That's most probably due to Volvos packing even more sensors : in addition to forward facing cameras, sonars and a radar (like Tesla), they also feature a lidar laser which is able to cover a few extra situations.
If you're going to try experiment with some dangerous new technology like adding a pilot-less hardware platform, better add it to a car that has already reliably been able to stop before an accident could happen.
As opposed to what's on offer safety wise with popular electric car brands :
- Tesla have also a lots of bells and whistles, but although they to get stellar security review regarding their survivability in case of crash, they still don't have Volvo's track record of avoid to get into a crash since almost 2 decade. Also Tesla is a direct competitor of Uber in the realm of startups aiming for pilotless drive.
- Renault Zoe are hugely popular electric cars in Europe (due to their very low prices). But currently, they're the only range of cars to feature NO options for FCAS at all. (Also Renault is currently working on experimental pilot-less platforms, and thus might be percieved as a competitor by Uber).
- VW has FCAS as a standard on its whole range, including even the fucking dead cheapest "UP!" cars (and thus also on electric variants like e-UP! and e-Golf), but ti's a much simpler and cheaper (Lidar only "City Safety"-style). (And also they're working on their own auto-pilot for their upcoming "e-" version rebirth of their clasical mini van).
etc.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
Well, then you obviously don't live in a cold climate. There are a few times it's happened: bought a second hand car and the engine cut out as I was accelerating onto the highway from a ramp - no power brakes, abs or power steering in the middle of a snow storm with ~2cm ice on the road. The second time I had to avoid someone who suddenly slid into my lane from an on-ramp (student, first time in a cold climate from California, she tried to brake but obviously that doesn't work on a frozen highway).
Neither times did anything get hit but we regularly have sheets of ice on our highway for weeks or even months at a time, especially the smaller 2-lane highways less often traveled.
Custom electronics and digital signage for your business: www.evcircuits.com
You're excluding yourself in that population
http://copradar.com/redlight/factors/
Of course like Trump you are better than everyone else.
What is your reaction time? The average human eye-brain is ~100-200ms up to 500ms in low-contrast conditions (like night driving) and then it's another 80-200ms to actually respond with a large muscle movement and then there is the time delay to actually have your leg move down the shaft.
Driver reaction times average ~2.3s in controlled environments although this could potentially be less in adrenaline-pumped situations.
Custom electronics and digital signage for your business: www.evcircuits.com
Probably because Volvo will accept liability for self-driving cars. https://www.forbes.com/sites/j...
a ride share program/app/company? It hasn't been about ride sharing in ages.
Many human drivers would plow right into the bag with block, but many others would notice it wasn't moving like a bag should and avoid it.
'Anything moving' fails in the fall. Unworkable. Also fails on litter.
We used to play _on_ the road. Shouts of 'car' would go up and down the street to let them past. Granted the older kids watched out for 4 year olds (and younger).
We already have, more or less, automated divided highway driving. That's the easy part. We might never have automated 25 mph side street driving.
I expect trucks to be rolling about 24x7 with single drivers soon. In virtual trains on long stretches, with the drivers (except in the first truck) resting, doing bible study, banging lot lizards, fapping, gambling (for a talented few, all at the same time) etc.
John McAfee 'It was like that time I hired that Bangkok prostitute; to do my taxes, while I fucked my accountant'
I've lost control going walking speed, just because I was trying to get around a car going the other way and had to drive out of the ruts. Do that when it's slippery and your car slides sideways, not much you can do about it.
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
This is a stupid argument. We're not talking about one driver, we're talking about possibly the same self-driving program in millions of cars. You ask millions of drivers if anyone has been in this position and you're going to get at least one. That's really all that matters.
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
Well, Lidar will need to go away anyway. That thing on the roof interferes with too many things, is too easy to sabotage (an empty soda cup will do, and you can expect some people doing that just for fun) and too expensive.
Lidar can also be spoofed: https://gizmodo.com/a-60-hack-can-fool-the-lidar-sensors-used-on-most-self-1729272292
Back to basics... your SDC needs to be able to not only detect obstacles, but also needs to be able to see them in context, meaning really understand what is going on the road. You don't need Lidar for that, cameras will do.
As for snow, the problem is mainly the falling snow, not the one on the road. The falling snow will stick to all surfaces, that includes camera lenses, ultrasonic sensors... and, unless your software really understand what's going on, look like a lot of moving obstacles. Same for loose leaves in combination with wind in fall.
Many human drivers would plow right into the bag with block, but many others would notice it wasn't moving like a bag should and avoid it.
Some might. Most wouldn't.
It's something that, if it were going to be a problem, would be a problem today. You might as well talk about shining high powered torches into driver's faces at night.
'Anything moving' fails in the fall. Unworkable. Also fails on litter.
Can we agree that there's a reasonable size difference between a child and the vast, vast majority of wind-blown leaves and litter we see in normal circumstances? If something the size of a child is blowing about, it's probably better to at least slow down to avoid it.
We might never have automated 25 mph side street driving.
"Never" is a really, really long time. Look at how far we've come in about the last three decades.
Assume a driver costs $40k per year. That's $960,000,000 per year for 24k drivers. Which means that this would pay for itself in a little over a year. Ok, this is a bit simplistic, but I doubt the real number is far off... Almost certainly not more than a two year break even point. With this point declining for the next batch.
And it's not just the monetary cost of paying the drivers. It's also the hassle of dealing with HR. Fewer (or no) drivers equals fewer headaches with unionization drives, sexual harassment of passengers, etc. etc.
It's the gift that keeps on giving.
Here, bus drivers have been striking for higher pay.
Personally, I'd like to see our council agree to whatever pay rises they want, on the provisor that whatever they give them, they'll also put into a fund that'll be used to implement driverless buses and trains.
Umm, except everyone is telling me autonomous cars are safer than people. Wouldn't the same be true for airlines and with that assumption, the airlines would be more likely to go with autonomous since the cargo is more valuable.
This is a stupid argument. We're not talking about one driver, we're talking about possibly the same self-driving program in millions of cars. You ask millions of drivers if anyone has been in this position and you're going to get at least one. That's really all that matters.
For PR purposes, yeah. Outside of that, all that really matters is that fewer people are killed via automated cars than they are via human-driven ones.
Aircraft already are heavily automated.
But the cost means it's still worth having a human pilot there. Just in case.
Also: Aircraft crashes are pretty amazingly rare. Sure, you've got a few, but it's only a few (no matter how much we freak out about them)
Even in 2014, which included a plane that was shot down and one that just vanished, we only had 761 deaths in that year.
As of 2010, there were 1.25 million car crash fatalities.
Seeing the difference?
Yes, it gets updated. By the traffic signs already in use that get put up or removed. The new rules are in effect immediatly.
Signs today are pretty standardised and very simple. There really isn't a huge amount you could do to improve them for automated cars anyway.
However, what I was replying to was related to GPS maps. You can go on about the fact that they need to deal with things as they are today. And I'd agree with you 100% - But the fact would remain that adding that data to a central DB is still a really good idea. For human drivers using GPS as well as for automated cars. Times have changed, and those doing this stuff need to change with it,
Well, Lidar will need to go away anyway. That thing on the roof interferes with too many things, is too easy to sabotage (an empty soda cup will do, and you can expect some people doing that just for fun) and too expensive.
People can already do some pretty damaging stuff 'for fun' - This is like those people who say that people will put balloons that look like rocks in front of the cars to fool them. That's already an option. However, by and large, they don't.
Reading the link you gave, I can't see how it's more than annoying.
As far as putting a can on top of it. I dunno, I'd assume that it would be noticed as soon as it was turned on.
For snow, I'd assume that these instruments could be heated.
Nope
Nope to them being heavily automated?
Not being a pilot, I couldn't comment. That's just what I found when researching.
Nope to the cost justifying a human pilot?
Considering the cargo/value of a plane and contents vs a taxi, I'd have to wonder.
Nope to crashes being rare? Yeah, they're pretty damn rare.
Nope to the numbe.......
Ohhh, I understand. You just don't see the difference. Sorry,there's a level at which I just can't help you.
It happens, I ran with a bunch of shitty kids. One was really psycho. Rolled a big old tire and rim into 4 lanes of traffic, about 5-6th grade. That's the one that got him busted. Anyhow...
'Litter' can be pretty large, but out of the way. The moving kind is generally small, not much bigger than a blowing bag or paper. Do AI cars even try to avoid cats in fall?
I have stepped on my brakes hard because I saw a child's toy (not a child) bouncing into traffic. I inferred the child I couldn't yet see. That can't be that uncommon an experience.
AI cars should stop for everything bouncing or rolling, at least. Which is trainable. But build an AI that identifies 'toys' in an image and runs realtime. (Coder/Training set builder at end of long 'sprint': Look for rolling wheels and call it a day.)
John McAfee 'It was like that time I hired that Bangkok prostitute; to do my taxes, while I fucked my accountant'
You should know that if you take a different route than Google Maps suggests you don't get paid for that extra time/mileage.
Uber uses Google Maps' API to estimate how long you should spend driving for a dropoff and pays you based on that.
In fact if you go off route a lot they will eventually suspend your account.
Almost every time I have had a dispute with a customer, Uber and Lyft have both ruled in my favor.
Of course I have a dash cam in my car so that probably helps.