Naked Science Forum
Non Life Sciences => Technology => Topic started by: jeffreyH on 19/03/2018 18:53:49
-
We are arrogant enough as a species to think we have it all worked out. Well if self driving cars so brilliant why has one of them killed a pedestrian?
https://mobile.nytimes.com/2018/03/19/technology/uber-driverless-fatality.html
-
As a species, we are arrogant enough to drive cars, even though we have known for a long time that we sometimes kill people that way.
The autonomous car doesn't have to be perfect- it just has to be better than us.
They are not there yet, but it's going that way.
-
How dangerous are self driving cars?
Already less dangerous than:
- Someone who is more interested in texting than in keeping their eyes on the road
- Someone who is more interested in having a good night out than in getting home safely
- Someone who has just driven for 8 hours
- The human driver supervising this Uber car who has not had to touch the controls for 4(?) hours
Often when there is a car accident, it is a case of "He says this, she says that, we can't tell if the light was red or green".
- And nobody really learns anything beyond what they already knew: Don't race a red light
- Self-driving cars keep a detailed record of the events leading up to an accident, so there is a good chance of knowing exactly what happened
- There is a good chance that this experience will be incorporated in future versions of the software, so they will improve over time.
In this particular instance, let's look for the verdict from the police about whether:
(1)the car was at fault
(2)the accident was preventable
(3)What happened with the human safety backup.
-
We are arrogant enough as a species to think we have it all worked out. Well if self driving cars so brilliant why has one of them killed a pedestrian?
Because if there's a driving fault with a type of self driving car, you can fix it and roll that fix out within a day or two to the whole fleet over the internet.
You can't apply a fix to all human drivers.
-
AFAIK self-driving cars rely on clear road markings that correspond to GPS maps, so work well on highways but have problems on unmarked roads and in cities where the GPS signal is distorted. Trouble is that these are the areas where pedestrians, animals and cyclists are most numerous and least predictable.
The average truck, taxi or bus driver will drive a million miles in a career without hitting anything, and the ordinary business motorist probably half that distance. And there are millions and millions of us. We tend to concentrate on the rare and spectacular without considering just how good humans really are.
I'm all in favor of autopilots but there are plenty of occasions where it's safer to disengage, and even more important to recognise when the AP has disengaged itself. Not a problem in a plane where you are at least 30 seconds away from anything else, but cyclists in particular have a habit of falling off, changing course, or appearing from nowhere in less than 0.5s.
-
Computer vision is a difficult problem, but no doubt getting better.
In this case, the reports are that the woman stepped out in front of the car. What was she wearing? Lights? Everything seems to indicate that using visible light, she would have been difficult to see, and likely would have been hit be many human drivers.
However, I think there are likely several things that could have been improved. For example passive IR might be good for night detection of life.
The car is supposed to also have various types of active imaging that isn't necessarily dependent on visible light. So, when was the woman first detected? Could she have been detected earlier?
Part of driving is to anticipate that others will do stupid things. So, perhaps slow down or move over to create a larger buffer around pedestrians, cyclists, and others that might cross one's path.
A lot of it is basic defensive driving. Not really a "fault" thing, but rather just rules to be safe on the road.
We'll see what the final report about the incident says.
-
I'd be interested to know the algorithm for dealing with animals in the road. Sudden braking for a dog is counterproductive - you are likely to be hit in the rear or skid on anything less than a perfect surface, whereas dogs have the sense to get out of the way. Sheep generally (a) don't have much road sense but (b) are still cheaper and softer than a rearend collision, so for most quadrupeds smaller than a horse, it's better to keep moving. Humans, on the other hand, are expensive to hit and completely unpredictable, and the law always favors the cyclist or pedestrian, whatever his state of inebriation.
-
in cities where the GPS signal is distorted
Apparently, the cost of GPS chips have come down to the point where sometime this year, some cellphones should be able to start accessing a more accurate timing signal.
This should give much improved location resolution, and much better immunity to multi-path reflections (eg when driving through city canyons).
Unfortunately, due to launch delays, these signals will not provide full coverage until around 2022.
https://spectrum.ieee.org/tech-talk/semiconductors/design/superaccurate-gps-chips-coming-to-smartphones-in-2018
https://en.wikipedia.org/wiki/GPS_Block_IIIA#New_navigation_signals
-
I saw the video of the accident, taken by the car itself.
- It was at night
- The pedestrian was in fairly dark clothes, wheeling a bicycle across the road (ie a tail reflector on the bike would have been facing the wrong way)
- There was a fairly short gap between the pedestrian becoming visible in the camera, and the car braking.
- But unfortunately, the car was still going too fast when it reached the pedestrian
- I didn't notice the car noticeably swerve to avoid the pedestrian
- The human backup driver looked up occasionally to look at the road, then back at something on their lap.
Some vehicles have lidar; they are expensive, but should work fairly well at night. I don't know if this vehicle was equipped with non-visual detectors like this.
I guess we wait for more detailed information and analysis.
edit: remove gender presumption
-
Yes, it appeared as if the car was driving with the lights dimmed and dipped to the right (right hand driving in the USA), with following distance on the one-way street close to the legal limit in the USA for high beams. The computer could have accurately calculated if it was "legal" to turn on the high beams. Dipping to the right makes objects approaching from the left more difficult to detect early.
Some older digital cameras had difficulties with low light situations, especially if moving. But, hopefully that is improving somewhat now.
I tried to lighten a screen cap just before the woman became visible, and got nothing. However, a digital camera should allow over-saturating shadow images, and then masking out the bright. Thus, multiple images may be able to do better than humans at "seeing" in a dynamic range of light/dark.
2s from visible to impact was short. No doubt image processing is not instantaneous, but there should have been time to brake and put the car into a hard left turn which would have avoided the crash. The advantage of a computer is that it should "know" which direction is safe to maneuver. I didn't see shadows of headlights from other nearby vehicles, but further analysis should tell whether a left turn was safe (although even hitting another vehicle may be safer than hitting a pedestrian).
-
Adding to Clifford's comments:
Visibility was definitely a major problem here - the pedestrian was crossing in a shadow, just after a pool of streetlight. The camera sensitivity may have been turned down for the streetlight, and missed the person in the shadow?
- A lack of dynamic range may have been a problem? CCD Cameras have a fairly wide dynamic range, but that range is typically compressed in software, especially if the dashcam was doing JPEG encoding.
- The light-colored shoes are the first thing to be illuminated by the downward-pointing headlights, then the slightly dark jeans.
- I imagine that a knitted dark sweater would be fairly effective at absorbing an Infra-Red LIDAR beam?
I wonder about the image classifier. Would a lone pedestrian have been recognised more easily than a pedestrian wheeling a bicycle?
As for evasive action, a sudden lane change to the left would have helped:
- swerving too violently risks a rollover (bad for pedestrian and passengers). SUVs with their higher center of gravity are not ideal here.
- swerving puts uneven pressure on the tires, and may impair braking (but today's antilock brakes are quite good)
- I didn't notice any headlights from cars in adjacent lanes. But knowing that the adjacent lane is empty (without looking) requires continual 360° situational awareness - not many human drivers achieve this. The iconic twirling LIDAR on the roof is ideal for this - but some manufacturers are now producing much smaller (& cheaper) units with a more restricted field of view that could be more easily incorporated into the vehicle body.
-
This article is a helpful, basic overview of the state of the autonomous vehicle industry (https://www.thenakedscientists.com/articles/science-news/uk-plans-get-driverless-cars-road-2021), which may be of interest to some readers.
-
Exotica, a British startup, is putting its cars to the test in Gatwick Airport
Every time I visit an airport, I think that the drop-off zone would be a very difficult place to operate a driverless taxi - there is total mayhem as (human) taxi drivers jockey for limited positions kerbside.
Of course, roadworks have major problems too... they aren't marked on maps, they change from hour to hour, and they (currently) have humans operating stop/go signs (although I've seen some automated ones too..).
Maybe these things will clear up as we discover that some jobs are just too dangerous for humans to perform....
-
The video was very dark, and it seems obvious that the car had no chance to stop.
However, there's problems with that explanation, the actual darkness is not that great, others have subsequently shown video at the same place and it's not really very dark at all, the camera seems to have been pretty bad.
And the Uber car has laser rangefinders that should have spotted the person. And with vehicle accidents, it's how fast you're going when you actually impact that determines the mortality rate. Apparently the vehicle was going at 38 mph before impact. If the vehicle had slammed on the brakes one second, if the road was dry, it would have been going about half that speed at impact, and the pedestrian would have stood a good chance of survival.
I mean, the pedestrian was clearly jaywalking but you're still not legally supposed to drive into them at full speed.
Anyway, there's a good Ars Technica piece about it:
https://arstechnica.com/cars/2018/03/police-chief-said-uber-victim-came-from-the-shadows-dont-believe-it/
-
A lot of interesting information has been coming out lately.
First of all, there are a few different companies testing autonomous vehicles.
Tesla technically isn't autonomous, but they've been adding sensors to theirs and what they call "autopilot" which is supposed to help keep a vehicle in the lane, stop for obstructions, and etc. But, they technically require/encourage drivers to keep the hands on the wheel.
Google/Waymo cars have been tested. Not without problems, but they're apparently driving about 5600 miles between "disengagements", or overrides by the human driver.
https://www.nytimes.com/2018/03/19/technology/uber-driverless-fatality.html
Apparently the Uber cars are somewhat lower, with one report suggesting that their cars have been going about 13 miles between interventions.
https://www.nytimes.com/2018/03/23/technology/uber-self-driving-cars-arizona.html
Both Google and Uber initially had two "drivers" in the cars, with one ready to take over, and one monitoring the computer system. Google reduced their assistants to one, and Uber followed, despite apparently needing more interventions. This likely led to split attention that was noted with the Uber driver.
Safety is ambiguous. It would be premature to say the autonomous vehicles are better than the human driver counterparts, especially if one counts the apparent periodic requirement for interventions.
There has already been one Tesla death in autopilot mode. And, now the Uber death.
The Uber Autonomous vehicles have driven around 2 to 3 Million miles, with 1 death.
On average, human drivers (USA) have about 1.18 deaths per 100 million miles driven. Obviously random events can occur anytime, but Uber is currently on track for about 33 deaths per 100 million miles.
-
Too dangerous.
Until they perfect the AI program self driving car are still dangerous.
And people will lose their job and can cause a massive chaos.
Better focus on other environmental stuffs.
-
Apart from the huge numbers of deaths on the road, if self driving technology appears then we'll llikely need far less cars. Cars embody about half the CO2 they're ever going to emit when they're first built. So if we can get more use from our cars, because they drive themselves around, then we'll need less cars, and the overall pollution from them will plummet.
-
Not sure about the arithmetic here. A car always and only goes where the driver or passenger wants it to. How would an autonomously-guided car reduce the distance from A to B, or the frequency of journeys? Shared transport along fixed routes is called a bus.
You are right about the energy input to make a new car. In the case of a conventional vehicle, it's about the same as the lifetime fuel burn, and in the case of an electric car, somewhat more. So if your primary concern is CO2 emission, the best strategy is to do nothing.
-
No, because one way to use self driving cars is as taxis. In that case a single car may be able to make as many as ten times as many journeys; equivalently you only need one tenth of the number of cars, and it's likely to be far cheaper than owning your own car.
In addition if the cars are electric, again you only generate about 1/10 of the amount of pollution while driving, so CO2 emissions are slashed.
-
No, because one way to use self driving cars is as taxis. In that case a single car may be able to make as many as ten times as many journeys; equivalently you only need one tenth of the number of cars, and it's likely to be far cheaper than owning your own car.
In addition if the cars are electric, again you only generate about 1/10 of the amount of pollution while driving, so CO2 emissions are slashed.
I'm not convinced that self driving cars will reduce much.
First of all, an EV only reduces CO2 if it is using power derived from solar, wind, hydro, or nuclear. If it is using power derived from coal, natural gas, & etc... the CO2 savings is minimal if anything. It is also more difficult to power fleet vehicles with electricity, as they essentially need recharged on the fly, whereas a typical single person commuter can be charged at home and/or at work.
Reducing the number of parked cars while maintaining the same number of active cars on the road will do little for traffic congestion, other than perhaps reducing the real estate required to park the cars.
The other thing that is inevitable is that driverless cars will allow children and people who would otherwise not be able to drive to start driving. Why won't your 6 year old be able to take the driverless car to school?
Then, the inevitable extension will be riderless cars. With your taxi service, the taxis will spend a portion of their time driving around without passengers. Likewise, why not just send one's car off to pick up the kids at school without "soccer-mom" driving?
For that matter, want a hot cappuccino, send the car to fetch it. Inevitably there will be services like grocery services that one could send one's car off in drone mode, then the the store will pick the order and fill up the car. Perhaps using drones to pick the items from the warehouse to load in the car. In fact, people like one-stop shopping for convenience, but with the drones, one could send the drone all around town to pick up the perfect items without worrying about the annoyance of going from store to store.
-
I'm not convinced that self driving cars will reduce much.
First of all, an EV only reduces CO2 if it is using power derived from solar, wind, hydro, or nuclear. If it is using power derived from coal, natural gas, & etc... the CO2 savings is minimal if anything.
That's not true. If generation is mostly coal, then EVs tend to be about the same as diesels, and still better than petrol.
But in the UK we already have a reasonably green grid, at night, we often exceed 50% nuclear and wind- with most of the rest being gas, which is half the CO2 output per kWh compared to coal. And the percentage of wind power is increasing year in year out, because it's now the cheapest option for power companies, and as the old equipment wears out, it is replaced with mainly with wind as well as other renewables.
It is also more difficult to power fleet vehicles with electricity, as they essentially need recharged on the fly, whereas a typical single person commuter can be charged at home and/or at work.
Not so much; there's already an electric taxi firm in Cornwall, they do one fast charge a day and then they top up overnight. And the batteries are getting bigger; if a taxi has a couple of hundred mile range; if the average customer mileage is 30 miles, then they can service 8 customers before they even need to recharge.
Reducing the number of parked cars while maintaining the same number of active cars on the road will do little for traffic congestion, other than perhaps reducing the real estate required to park the cars.
The other thing that is inevitable is that driverless cars will allow children and people who would otherwise not be able to drive to start driving. Why won't your 6 year old be able to take the driverless car to school?
True, but you could arrange a group pickup for multiple children, in many cases that would be both cheaper and easier.
Then, the inevitable extension will be riderless cars. With your taxi service, the taxis will spend a portion of their time driving around without passengers.
Yes, the software can solve the Travelling Salesman Problem and arrange a minimal route.
Likewise, why not just send one's car off to pick up the kids at school without "soccer-mom" driving?
Yup, that's basically a school bus.
For that matter, want a hot cappuccino, send the car to fetch it. Inevitably there will be services like grocery services that one could send one's car off in drone mode, then the the store will pick the order and fill up the car. Perhaps using drones to pick the items from the warehouse to load in the car. In fact, people like one-stop shopping for convenience, but with the drones, one could send the drone all around town to pick up the perfect items without worrying about the annoyance of going from store to store.
Again, a Travelling Salesman Problem, and you already have that, Tesco's already deliver cappuchinos and other groceries.
-
Increase convenience and access... and I just don't foresee people carpooling more and driving less.
School buses, yes, but I'm not quite sure why so many people prefer to drop their kids off at school and pick them up. Some due to not going to the closest public school. Some because the buses typically cut off services for less than 1 mile from school, and 1 mile is too far for a kid to walk or ride a bike. After school activities? Just want to "be there for the kid"?
We may initially see taxi services adopting the driverless cars, but I can't imagine that private sales or leases will follow very far behind. Most of the major manufacturers are investing heavily in accident avoidance systems, but full autonomous mode is only a few steps behind.
-
The thing is, if you own an autonomous car, running it as a taxi means the entire car is a net profit. So we go from a situation where a car costs you money to own, to one where it pays for itself. That means the number of taxis goes up from a tiny percentage of the vehicles on the road to potentially the majority.
So you don't wait for a taxi, it turns up on time, from a MASSIVE fleet of taxis, ordered over the internet. That also means that the cost of taxis-and cars in general- is plummeting, together with what is likely to be a large improvement in the CO2 impact due to their being less vehicles, and from electrification.
-
No, because one way to use self driving cars is as taxis. In that case a single car may be able to make as many as ten times as many journeys; equivalently you only need one tenth of the number of cars, and it's likely to be far cheaper than owning your own car.
From Alan's Primary Reading Book
A nice guy called Wayne lives in the next village. He has a big car. We phone Wayne. He brings his big car our house and helps us put our cases in the big car. Then he takes us where we want to go. We go to the airport. He meets other people at the airport and takes them where they want to go. Some of the people are very old and need a lot of help. Some of the people don't speak English but they have a piece of paper with hard words like Leicester, Gloucester or Irthlingborough written on them. Wayne can read and write (and speak Spanish) so he knows where to take them. He can stop at a café or toilet of they want to. Wayne is a taxi driver.
-
With normal human-driven taxis you're having to pay for the person's time of course, and traditional taxis will usually only work part of the day, whereas autonomous ones could go 24x7 stopping only for recharging.
We'll see how it plays out, but the interface will probably be an app, in your own language, on your own smart phone. If you need help there's already porters at the airport that can help you, and the taxi company (which your car may be working for) or carers can provide people to help load the vehicle at the customer's end and could be dropped back off- en route, or take different taxis, rather than riding all the way with the car.
-
Not particularly relevant, but as a matter of fact most "black cabs" are rented or owned by a syndicate, and actually do work 24/7. Wayne's business partner Julie likes night work - usually one-way for extra pay. They also clean the car between trips, and hold up placards at airports and suchlike.
"Porters who can help you" is not idiomatic English, though some Americans remember when it had meaning.
-
One of the reasons people like their own personal cars is that they'll leave personal belongings in their car. Whether it is leaving a book in the car, prescription sunglasses, coffee cups, etc.
Sending one's car out for taxi service would require at least some sanitizing of the vehicle which some people would embrace, others wouldn't.
It may be that Uber would put themselves out of business if this comes to fruition. How long until 100 different companies build their own car-share websites? Perhaps with less overhead or various other improvements over Uber.
Another issue, both customers and service providers will have to have a method of ranking each other. So a customer trashes a car without supervision, and they'll quickly get shunned from the carshare community. Likewise, people who send out trashy cars will also get shunned.
FLAT TIRES? Now, some companies are coming out with drive-flat tires, but there will have to be provisions for driverless maintenance, especially if running a taxi service.
-
You will hear crash news of those self-driving Tesla car, EVERYDAY. That's how it is. It's still under-development so don't expect much better from them at this moment. Yes after 10-20 years we can be assured they will be safe but not now.
-
This is a bit of a half empty thread. Since my original question was 'How dangerous are self driving cars?' It could equally well have been phrased 'How safe are self driving cars?' It would have been interesting to have seen how this affected the types of responses.
-
Either way, the answer is 64.
-
You will hear crash news of those self-driving Tesla car, EVERYDAY.
So, the crash of a self-driving car is newsworthy (although I've only heard of a few in the past few months, not every day).
Does this newsworthiness demonstrate that they are:
- Incredibly dangerous? (But then I do hear about the crash of human-driven cars every day)
- Incredibly safe? (The fact that one crashed is newsworthy)
- Something new, which automatically makes it news!
- Something feared? (The computers are taking over this, too)
- Something that only the rich can afford, so I am fascinated by their misfortune?
The truth is probably a mix of all the above. What matters to me is:
- Are they safer than an average human driver? If so, they are a good thing.
- Are they cheaper than paying for a human-driven taxi? If so, they will take over.
- How long will they take to reach this superior state?
-
@CliffordK quoted statistics a few posts ago that show selfdriving cars currently kill about 30 times as many humans as human-driven cars. Not sure what counts as newsworthiness, but no journalist would let the facts get in the way of a good story, and no politician would let them get in the way of technological progress.
-
We are arrogant enough as a species to think we have it all worked out. Well if self driving cars so brilliant why has one of them killed a pedestrian?
Arrogant or not, cars in itself is a 'old' paradigm for travelling. Self driving cars? Might as well put billions in attaching a bunch of rockets on a camel.
-
Tell that to blind and older people that can't drive themselves around.
-
One of the problems of today's self-driving cars is the difficulty of making a smooth transition from automated mode to manual mode, when the vehicle finds a situation that it cannot handle.
This particular vehicle has limited capabilities in self-driving mode, but has put a lot of effort into the user interface:
- Ensuring that the driver is aware of the road situation, even if they are not currently in control of the vehicle
- Smooth handover from human to machine
- Smooth handover from machine to human
See: https://spectrum.ieee.org/transportation/advanced-cars/2018s-top-ten-tech-cars-cadillac-ct6
-
Self-driving is an inevitable trend in the world. But I am afraid of taking an automatic car in the future. :-[
-
Whatever you may think, here is the opinion of the police and courts:
https://www.msn.com/en-gb/news/uknews/shocking-moment-tesla-driver-turns-on-autopilot-and-moves-to-passenger-seat-while-driving-on-motorway/ar-AAwt2qV?MSCC=1525037636&ocid=spartandhp
-
That's not a self-driving car, it's a car driving under autopilot. Would you leave the cockpit when your aircraft is solely flying under autopilot?
-
How safe are human drivers? Are they getting worse as they spend more time being distracted by phones (such as a lorry driver who ran into a line of stopped cars on a motorway while he was trying to find an mp3 track on his phone)?
I'm more worried about human drivers now than self-driving cars, but I also object to self-driving cars which aren't properly safe even if they're better on average - they need to be better on average in all situations. We've seen a major fault with an UBER self-driving car where it failed to slow down when its lidar must have picked up a pedestrian walking into its path. If that fault repeats in daylight when a human driver would have done a better job, that system isn't adequate.
But the worst problem with self-driving cars hasn't been addressed yet - they could be used as bomb delivery systems.
-
Just like ordinary cars and trucks nowadays. Except the police are more likely to stop a self-driving car if there are no visible passengers. And idiots are cheaper than computers.
-
Indeed, self-driving cars can be considered as a hell of innovation, but the technology is yet not even remotely close to being perfect. A friend of mine has gotten himself a fine Tesla Model S with an upgraded self-driving package; however, it turned out that it is not really safe to drive in this mode. Not yet, at least. For this particular reason, I am sticking to my beautiful hyundai grand i10 as for now, haha.