Sometimes you need a human driver

• Will these self driving cars have an algorithm that makes them run red lights in seedy neighborhoods?
  
  What red lights? This is the part that people don't seem to understand. In a autonomous car platform... there would be no red lights. No green ones either. All the money needed to keep traffic lights up and running would go away. Idling while not moving would be a thing of the past. Waiting for grandma to notice the light changed would no longer be a thing.
  
  Don't we need to replace failed GPS satellites? I thought I read somewhere that we have fewer now than we used to have.
  
  GPS is only necessary to drive cross-country. Google's current generation of cars is able to do independent reckoning based solely on Google's Street View data. This is very important for tunnels or large cities where GPS becomes very unreliable.
  
  Also, GPS isn't used for short-term navigation. Loosing GPS will never make their car hit a kid, because it's using cameras and radar for that.
  
  How does it account for terrible weather conditions or bad drivers?
  
  It wouldn't take much for automation to handle weather better than most drivers do, after all it could use a database of previous scan by other automated cars that have taken the route to know where the road is.
  
  As far as bad drivers go, it will be able to see much smaller deviations in behavior than people would generally perceive and likely be able to anticipate a bad drivers maneuvers just by monitoring nearby cars wheel orientation, changes in center of mass by defection, etc. The same thing can be done for pedestrians.
  
  The technology isn't 100% but it doesn't have to be, it simply has to be better than human drivers, honestly that isn't a high bar given how common distracted driving is.
  
  The issue is that the sensors used to notice obstacles and localize the car (specifically the LIDAR) break down when the surfaces around the car are highly reflective. That tends to happen when everything is wet. The computer can't drive if it can't figure out where it is in space or what's around it.
  
  I'm an engineer working on a university project that uses radar to do 3D real time imaging much like LIDAR. The resolution won't be even close to that of LIDAR, but the frequency is low enough that it should be able to see through snow, rain and dust. Rain is actually harder than the others because liquid water scatters better, but at our frequency (~6-11 GHz) it should be acceptable. It's still kinda early to see how it will all go, but we are hoping it will compliment LIDAR in self driving cars. We are planning to open source or publish as much of the hardware and software as possible by this summer. Cool stuff though.
  
  When you consider hilly roads and icy dangerous conditions consider that if self-driving cars were the norm and one of them hit an icy patch, EVERY OTHER CAR would know instantly that the section of road is icy since the cars would likely be able to easily share road condition information with each other (using today's technology even, not considering whatever type of networks come next). So gone would be the days of the multiple cars careening through an icy intersection or sliding down black ice on a hill.
  
  stopping for steam won't cause more accidents though. cars stop suddenly for various reasons all of the time. the cause of the accident is drivers following too close to react. that steam could have been a squirrel or a skateboard.
  
  speaking of squirrels, i saw a car leave the road at 60mph to avoid a chipmunk. I will have much more faith in software if the emergency response is "come to a controlled stop" rather than "EMERGENCY MANEUVERS AT HIGH SPEED" that human evolution has developed.
  
  Are these systems perfect? No. Are they better than humans. YES! by orders of magnitude.
  
  A lot of people think that anything less than perfect is unacceptable. I'd agree with that if it wasn't for the fact that ~40,000~ people die every year in the US in auto accidents. That's deaths, not accidents. That giant number doesn't even count lost work, totaled vehicles, broken bones, pain, or even lifetime disability.
  
  I find it really interesting that if somebody said "Autonomous driving will never be flawless and at best, will still result in 1,000 deaths per year if it completely replaced human drivers" - I admit that sounds scary.
  
  Rephrase that as "if autonomous vehicles reduced traffic deaths by 98%, would you still feel safe with human drivers?" and it doesn't even sound like a question. IMO, the "I feel safer when I'm in control" argument sounds a lot like "elevators are scary. Lets go rock climbing!"
  
  At some point the cars could even coordinate actions, so the car behind knows the other car sees an obstacle and is breaking even before it's own sensors can detect either the obstacle or the deceleration of the car ahead. And after that I'm looking forward to slipstream technology; with cars teaming up and forming efficient synchronized chains of cars in close-distance for best aerodynamic effect.
  
  People who ask too much of initial technology always get me down. The car. Drives. Itself! Holy shit, man! Just accept how awesome that is!
  
  So the controls on a self driving car which is mass market need to essentially be a "stop" button. As you point out, we can't trust people to pay attention for events which happen once every 3 months - the car must be able to identify when those are an issue and handle them on it's own. Another option would be for the car to force the driver to deal with them. Either way, the car must be good enough that it doesn't need input from the driver. The cool part about these things is that they are capable of learning from the entire fleet. Every hour of driving training adds an hour of experience from every car in the fleet TO every car in the fleet. If you have 100 cars on the road, each week you add maybe 2500 hours of driving experience to your fleet. Machine learning is pretty awesome that way - just need to wait for the training period to end.
  
  This scenario: you are asleep in your self driving car as it approaches a tunnel through a mountain. As it reaches a tunnel, a child chasing a ball runs into the road and stumbles. The car is too close to brake, the only option to avoiding hitting the child is swerving, causing the car to collide at full speed with the mountain side, killing you. Who's life should the car choose to save?
  
  If we are talking about self-driving cars, then most often that means that a corporation of some sort owns the car and the user pays a fee for the transportations service. If the corporation programmed its vehicles to choose the victim over the human "cargo" in the vehicle, the corp would soon be in some serious PR nightmare, since nobody would want to use their product if it was certain that in the case of an accident, their life matters nil to the service provider. Furthermore, if the world continues on the same path of capitalism, then corps will continue to only care about the bottom line, which is income. I don't know any business which favours the death of its customers. Loss of customers = loss of revenue, which leads to discontinuation of the service. Therefore, the self-driving cars will choose the safety of the cargo over the safety of the accident victim.
  
  To expand on your 'convenient driving deaths' comment a bit - also the ability of the self-driven car to make "sacrifices". By this I mean one car foresees a major accident will occur two minutes in the future unless they prevent the accident somehow by essentially becoming a kamikaze car to kill one and save many.
  
  That would be a huge flaw in the programming of the cars if it essentially is programmed to get into an accident. Maybe I'm being more idealistic about how well these cars will work in real life situations but I would think that if the cars communicate with each other so every car knows where the other will be in the future then I would think that they could communicate enough to easily change speed or direction to avoid the collision.
  
  To me it seems like a pretty big leap, so maybe that's why I didn't get it as a rhetorical. "If something can't deal with every life-threatening situation, we don't use it" is not the standard we have usually seen required to adopt a new technology - just that it decreases the loss of life with its introduction. Chemotherapy might be a relevant example as something that has inherent issues but saves many lives, and is therefore implemented.
  
  When people die from mishaps with new technology, history tells us that lots of times it just means that that technology will get safer. Cars used to not have seatbelts and had hard dashboards. This led to unsafe conditions. Cars were made safer.
  
  Also, driving currently is relatively dangerous all ready. Thousand of people died from car accidents in the US. In 2012, the number was 30,040. If google made a car that replaced all driven cars and ever year 10,000 people died it would still be three times safer than what we have currently.
  
  Do you think we will say goodbye to all that technology because of an accident? Of course not! On the contrary, with every accident it seems (rightfully so) that our answer is to add more technology. There's a reason that cars perceived by the public as 'safest' like Volvos have the highest uptake of technology. Both in passive safety features as in active overruling of human judgment by the car itself (the essence of self-driving cars) And it's not just perception either. Every single statistic shows that the constant reduction of grave injuries and deaths in car accidents can is also correlated by new active safety features. Volvo is famously aiming for no fatalities with their new cars after 2020. And they are looking to technology to make that happen. So I think the more likely answer from the public under pressure from continuing road accidents will be a demand for more, not less technology.
  
  Automatic cars will definitely fail at some point. That's going to happen, but it is not failure that makes a thing banned. It is repeated failure. Planes crash, but we still fly in planes. If 99% of all planes that took off, crashed, we wouldn't fly in planes anymore. So, there is acceptable risk. You might die when you get in your car, but it's an acceptable risk for you.
  
  That might fly in Boston or Manhattan or San Francisco, it ain't gonna fly in Wyoming, Maine, Utah, Arizona, etc. And good luck in Texas.
  
  My only question is what happens when the grid goes dark and no one knows how to drive? Has science fiction taught you nothing?!
  
  That 'argument' can be used against all kinds of technology, but that doesn't stop you from using technology, does it? Why should it be any different in this case?
  
  I would imagine it wouldn't be too different than if the grid goes dark and no one can get gas. Imagining worst case scenarios doesn't seem like a good enough reason to justify ignoring all the potential benefits.
  
  the first time someone gets killed in one of these they will be sued off the planet ...
  
  Will the first death be an outrage to the public, yes. but from that they can make improvements - but 32,000 people died in motor vehicle accidents in 2013? where is the outrage there? we can go on forever trying to improve the vehicle safety feature. but a human is still behind the wheel.
  
  got to be one of the dumbest ideas ever imagined, not doable as you have to have a socieity that wants this, trusts this and economically sees it as it's best choice ... non of those qualities are here, what an arrogant company
  
  And I guess the airlines were arrogant forcing these jet airplanes down our throat when good old steam trains were just fine taking 2 weeks to get us across the US? Technology shapes the world and it will never stop.
  
  The problem of a snowy pass is a good example of misguided thinking. An autonomous vehicle has reaction time many times faster than a human, it might not even need snow tires. Even if it did, the worst case scenario is you hire two guys to man stations on each side of the pass and just fit and remove chains all day. It seems like the idea of partially automating the job hadn't occurred to you.
  
  Automated Trucks
  
  Trucks can back themselves up now. Heck, most regular cars have all the required equipment except for the automation to be able to back themselves up. My truck knows when I'm about to hit something while backing up, if it could control itself it would have no problem backing up. There are cars that can parallel park themselves, more complicated than just going in reverse. And chocking tires could easily be replaced with an automated pylon of some kind that would hold the trailer and lift it off the tires. And a truck made to be automated would obviously have an automated way to disengage from the trailer it is hauling. Then it would proceed to go to another trailer and attach to that one.
  
  It sounds like you're just taking non-automated trucks and saying they can't be automated instead of considering that an automated truck would be a new thing, with new features and ways of doing things.
  
  The automated truck is proof of concept not proof of real world application. John Deere had this down 15 years ago yet not every piece of industrial, forestry, agricultural, or lawn care equipment is automated. All of those things require a lot less interaction with other humans and their proof of concept has barely been implemented (pardon my pun).
  
  I'm not arguing that this is ready for prime-time. I'm not arguing that every job a trucker does is going to be automated. I'm saying that when this technology is ready for prime-time (which is going to be soon) the cost of changing the industry to meet the needs of autonomous vehicles will be negligible compared to the saving from doing so. Your example of John Deere just goes to show what happens when the savings from automating are not large compared to the cost of doing so.
  
  They don't 'have' to unload it. That is just what they are expected to do. In reality, anyone with an ounce of brains could unload it. If trucks are automated, things are obviously going to change in that regard, because if there is no driver in the cab, you will need to have someone on site to unload the trailer. Just because it may be common to do the unloading one way today does not mean that this will persist in the future.
  
  If the front tire blows on any truck, it could kill someone. It does not matter whether the vehicle is automated or not. The truck will likely be programmed to instantly recognize the error condition, and take the countermeasures statistically most likely to result in the safe deceleration of the vehicle.
  
  Sorry, I disagree with your statement that automated trucks are not going to arrive within the next sixty years. Humans, for the most part, are terrible drivers and as soon as technology companies can demonstrate that automated vehicles are safer, on average, than human-controlled vehicles, automation will come at a rapid pace.
  
  Automated trucks will take away jobs because companies are not going to invest in an expensive technology if they cannot save a lot of money. There might be new jobs created, but it will be unlikely to equal the number of jobs lost. Why pay for someone to sit in a cab for 12 hours a day once automation can drive trucks more safely than humans?
  
  Look at trains for an example. A subway train used to have an engineer and a conductor. Then newer systems like BART automated the entire process and made it so that the engineer can do both jobs. Newer, fully enclosed stations can have trains that can be fully automated without the need for anyone on board.
  
  I suspect trucks will be much the same. At first, automation will be like autopilot or already existing crash-avoidance systems. Once the autopilot can prove that it can actually drive a truck much more efficiently and safer than a human, the human will be eliminated, except in emergencies, when operation of the truck will be handed to a driver at a remote dispatch center.
  
  I love road trips and I know there are a lot of good truck drivers, but I see so many of them doing things dangerously and are the cause of a lot of stress on these road trips. I don't really care how many jobs this costs. Bring on the automation. There is absolutely no way that autonomous trucks, taken as whole, will be anywhere near as dangerous as manned trucks when also taken as a whole.
  
  The same goes for the average driver in a car honestly. Too many distractions. Just let the robot drive and you can text away to your heart's content.
  
  As a person that works in Transportation and handles crash statistics daily, humans are simply awful at driving, and people that think they aren't are simply more dangerous because they are wrong; car related deaths have gone down over time because of tremendous safety engineering, both on vehicles and the road, not because people are any better at driving. The reduction of road fatalities would be more in the 90-95% range.
  
  Should your self-driving car be programmed to kill you if it means saving more strangers?
  
  Can it be programmed to kill as many strangers as possible? That's a much more relevant question to me.
  
  Can it be programmed to run over cats? That's a much more relevant question to me.
  
  If it doesn't put me first, then why would I buy it to begin with?
  
  Fark your 'self driving' car and fark you. My right to freedom of transportation will not be taken from me by Google and their Obamacronys. (Ow! That was my feelings! gets quite upset about these things)
  
  If some nitwit with his nose jammed into his smartphone walks in front of me the car better as hell be willing to run his dumb ass down.
  
  Just what do you think you're doing, Dave?
  
  If self-driving cars swerved for pedestrians, teenagers would discover this fact in no time and have a new way to occupy themselves when they're bored by running cars off the road. So it's important that self driving cars run down pedestrians at all costs.
  
  Yes, I think that self driving cars should be programmed to save as many lives as possible, even if that means my car may choose to kill me to save others.
  
  The way I see this is that if I'm driving a car that may choose to kill me to save others, that's a risk. But it also means that all the many other cars that I'm interacting with may choose to kill their owners to save me.
  
  Given that I only have one car and interact with many, the result is a net gain in my safety. And it is reciprocal, resulting a net gain for everyone.
  
  I don't care what my self-driving car does, as long as it allows me to drink heavily while it's driving me somewhere
  
  In those situations, the car should be programmed to kill us all and let Robot God sort it out.
  
  As I say in every one of these threads, the situation would be so rare that it wouldn't be worth trying to explicitly account for. I'm fine with the car choosing the least bad non-suicide option.
  
  So the cars will do for driving what autocorrect does for typing?
  
  Go duck yourself
  
  The software designers estimation of whether swerving for more room to stop is better than slamming on the brakes. The vehicle will likely be programmed to do everything it can to maintain control and avoid impacts.
  
  It's incapable of making moral decisions. If it swerves to avoid a busload of orphans, and that swerve sends you hurtling off a bridge, it's likely because it didn't understand that it was on a bridge and put avoiding collision higher than running off the side of the road. ...Of course since it won't be speeding or driving inebriated, it will see that bus coming from a quarter mile away. It has no blind spots. Said bus would have to swerve erratically into the cars lane, at which point what could anyone do?
  
  You might overreact and drive yourself off a bridge or hit the brakes and spin out of control. Since this is a computer driver it won't overreact. It can maintain precision braking and steering up until the last millisecond.
  
  The odds of it avoiding a collision in an otherwise impossible situation are far better than those for a human. The more cars are wired together, the more of a total ass a human driver has to be to catch them off guard.
  
  I can't wait for old people who should have had their licenses revoked a decade ago to get cars we don't have to worry about them driving.
  
  Self driving cars won't make a choice between hitting someone and killing you they will attempt to avoid collisions like a human would but be much better at it. None of these ethical issues are real they are just the imagined bullshiat of people who don't understand the technology.
  
  People are not expected to take control, they are allowed to take control. The cars will be more affective than humans in almost any scenario you can think of. The steering wheel is there for people who want to drive the car themselves, it will not be implemented as an emergency measure. In no way can someone possibly react quick enough in an emergency situation with their seat turned around and the wheel pushed in.
  
  
  
  
  
  Could it deal with this?
  
  
  
  Well, could a human driver?
  
  All I can say about self driving cars is this : itll never happen in places where snow falls. I'm from the Niagara NY region. There's a blizzard one day and is icy the next then 65 the day after then blizzard again. Variables like snow tires is the biggest factor and I don't think a computer can predict road variables like a winter Strom with black ice below 5" of new snow.
  
  It may not happen right away but we'll see if ever. In the mean time drive manually.
  
  So you think that a computer that knows:
  The temperature of the roads
  The moisture content of the roads
  The traction ratio of the road and tires
  The wind direction and speed
  What is around the vehicle, 360o , for well over 100m
  The depth of the snow on the road
  The density of the snow on the road
  The amount of treadwear on the tires
  The types of tires on the car (right down to size, brand, and model, not just "snow" or "all-season"), and all their specifications including the hardness of the rubber and the tread pattern
  The amount of pad wear on the brakes
  The wheel alignment, and how far out-of-spec it is.
  And can calculate the safe stopping distance down to an inch, correct for skids, pump the brakes when needed (far better than ABS and TCS can now), and everything else, can not drive as well in the snow as your meatbrain?
  
  Someone's a little overconfident.
  
  Self-driven cars will definitely SLOW THE FUCK DOWN for winter driving. People who drive, don't slow down, and that's why accidents happen. For the record, I'm from Rochester and I support self-driving cars because of the constant, daily clusterfuck that happens when people accustomed to flouting speed laws expect to keep on speeding with bad road conditions.
  And I also support it because I'm a bicyclist, and self-driving cars will respect me in a way that drivers don't.
  
  You really think your intuition will be better at handling inclement weather than a computer that's receiving data from dozens of sensors thousands of times per second and adjusting its outputs via feedforward-feedback systems? Dream on.
  
  You will be many orders of magnitude safer in an autonomous car in any set of conditions.