> By early 2015, Fairfield thought they were getting close: The cars were clocking full days at Google’s testing grounds without needing human takeovers. They had driven 1.2 million miles on public roads and could pull over for emergency vehicles. They could detect and brake for squirrels, and read hand signals from cops and construction workers. They knew when to honk politely (Just making sure you see me here) and when to blare it (You’re about to slam into me!). Then the team spent the next year putting another 800,000 miles on the fleet’s collective odometer, to fine tune everything. And it gave 10,000 rides to employees and “guests.”
There are good reasons why things like these happened:
> The law of some states, such as Louisiana, explicitly lists a killing in the course of defending oneself against forcible entry of an occupied motor vehicle as a justifiable homicide.
While I'm interested in self driving cars that still feature manual controls you can override, like Tesla, I have no interest in a future controlled by Google's AI without my input.
It's more that if an autonomous car with no human stalls, due to a traffic light that's stopped working, or an oddly shaped plastic bag in the road, it's stuck, and would cause a traffic jam. Not sure how to resolve such a scenario, can't rely on someone remotely taking control, as mobile data doesn't have 100% coverage, so a human who is allowed access to the car would somehow need to get thorough the traffic jam to move the vehicle on. Can't think of an efficient way to resolve this situation.
I mean the other day on my daily commute there were two cars right in the middle of a freeway, one attempting to jump start the other and they were at it for a while. This is not shockingly new ground.
If google actually starts selling a self-driving car and doesn't give any information about whether it can deal with a policeman-controlled intersection, then you can be concerned. Until then, implying their self-driving car programs haven't considered the basic problems that any uninformed commenter on the internet can think up is not incredibly helpful.
I can guarantee you that on a one to one driving skills shootout between you and the software you will loose 98% of the time.
In any case, I'm not talking about whether it "should be allowed to drive", I'm talking about the level of capability an autonomous car would have to reach in order to become a viable product.
Hell yes I will still be using a "non-autonomous transport appliance" in ten years - it's called a "motorcycle" :) - and I don't want your high-tech self-driving future if it has no room for that.
People will continue to spend stupid amounts of money on hobbies. I'm not going not hand in the keys to my Landcruiser or my CRF250. But I don't leave those parked at the train station as a rule either.
That remark was separate from and unrelated to the actual point I've been trying to make: there is a very significant gulf between the notion of a "self-driving car" as a form of automation built into a vehicle which is still under the control and responsibility of a human driver, and the notion of a "self-driving car" which is a fully autonomous robot built in the shape of a car, that must be capable of doing something reasonable in all situations without depending on a human. The latter is a much harder problem than the former, and we don't yet know whether it can be solved in a way that yields a practical, profitable transportation system of the sort many people like to fantasize about when they discuss (for example) Google's autonomous vehicle program.
It is very clear that autopilot will be a successful feature and we can reasonably expect continued improvement in human-driver-assist technologies for years to come.
But people keep on handwaving past the very important gap between "automation which assists a human operator" and "autonomous vehicle which gets around on its own", as though doing well at the first problem necessarily makes the achievement of the second problem inevitable.
Driver assist technologies make all kinds of sense but they do not fundamentally change the nature of our transportation systems. They just make it easier to do a better job as a human driver.
Getting to the point where most wheeled vehicles are self-piloted autonomous robots with no human driver interaction...? I still think that's a really hard problem.
Sales. That's what makes a viable product. One that will sell. Despite elsewhere in this comment chain you saying "I'm talking about product viability not technical" I can't help but think that your response to me was about technical viability.
People are already willing to buy and trust their cars to drive for them right now, when they're specifically being told "don't trust it to drive for you". The second it's legal to advertise "let it drive for you!" autonomous cars will sell regardless of their ability to perform well in unregular circumstances.
But people keep talking about "self-driving car" as though it is going to be a robot capable of running errands on its own, with no human driver and not even necessarily any human passenger. The question is still open: can this be done well enough, enough of the time, solving a valuable enough class of problems, that producing such car-shaped robots will actually become a significant industry? Furthermore, will the problems solved by such hypothetical car-shaped robots be significant and widespread enough to justify the hype currently being lavished on them?
Personally I'm of the opinion that because we've been making tools for 2.5 million years and don't seem any happier than the average animal, the very concept of a new tool making all that much of a difference is absurd.
Forget it, driver-less s cars will be the norm in the future. You won't be able to afford the insurance associated with a meat-driven car.
If we're crediting the car with being able to handle merging into a crowded multi-lane highway and dealing with idiot human drivers who don't indicate or hold lane position very well, it seems a bit odd to arbitrarily presume that it won't be able to identify non-hazards like said tumbleweed.
Does it drive better than me driving while being supervised continuously by a computer? I still make the important decisions, and my mistakes are pointed out by the computer.
The competition for self driving cars are not humans, but humans being adviced by driver assistance programmes
https://www.wired.com/2015/07/hackers-remotely-kill-jeep-hig...
Are there seriously cars on the road with complete fly-by-wire brakes which have no pure mechanical backup? That's horrifying. How did we go from dual redundant hydraulic systems that work (if in a reduced capacity) even when the ignition is off, to something at the mercy of some code that can be bamboozled over 3G?
For example getting out of the way of emergency vehicles on a blocked road. It will ask you what to do and if you tell it to drive up the pavement or into a bus lane or cross to the wrong side of the road -- the car will just do it.
I do wonder about all the non-verbal human-to-human communication. But a friendly baxter type screen with eyes that move (and messages that scroll across) would solve that.
We have to remember that these cars are all feeding unusual situations back to the google BRAIN that is learning. It'll watch us react to an emergency and learn. It'll see that 20 or 30 times a day and generalise a response.
The fully-autonomous route runs head-first into the inevitable, messy problems with software: security, maintenance, feature creep, complexity, unreliable networks etc. In theory, humans could nail all of these in car software. In practice, it's highly unlikely. The software industry can't even get basic security for IoT devices right.
The human-in-the-loop route runs directly into the only problem more daunting than reliable software: human nature. Human overrides for self-driving cars that need to be activated in a timely manner will not work. The drivers won't be paying enough attention to react in sufficient time. Note that overrides add even more complexity to an already fantastically complex system.
Say an always on self driving car, but it doesn't drive. Human has to drive, and the computer can correct any obvious mistakes the human does. Keep a really high liability for the car manufacturer though
ED: Not hitting kids is manditory, not blocking traffic is mandatory, parking in every random stop is optional as long as their are safe places it can drive to.
Basically, you don't need a perfect solution in every case, just a safe solution in every case. A self driving car that get's to a gas station is a good enough step 1 even if you have some human attendants there to fill up. Sure, it might add $2 bucks per fill up that's still cheaper than highway driving.
PS: Self driving car, only works in NYC is also useful.
Imagine I am a cyclist and I don't signal but I swiftly swerve to the right. Unless the self-driving car is already at a slow speed, it would impose an undue risk. Or, it is hard to expect that humans will rigorously adhere to this traffic discipline.
This is why these cars will drive slow for the foreseeable future and will frustrate a good chunk of the drivers.
Unless it gets dedicated lanes, the way bicycles do.
These cars are overhyped. They've been just around the corner for the last 10 years. They are the Duke Nukem Forever of Silicon Valley.
Not really, they're correctly hyped. The concept is revolutionary.
> They've been just around the corner for the last 10 year
B-S. Holy revisionism BS!!
We were NOT talking about self-driving cars in 2006! There was no company seriously designing fully autonomous cars in 2006! It was fully in the realm of science fiction, no investment, no major players, nothing.
And you make it sound like it was supposed to be released 10 years ago! Vaporware! They told us we'd have AI cars in 2006! It never came! <---- False reality that never occurred.
Why do you feel the need to lie about recent history? Shouldn't you observe that if you have to make a nasty lie to support your argument, maybe your argument is wrong?
You're definitely parroting the contrarian point here, but you're being very low-effort about it. Contrarian can be cool but please try harder.
I don't know how serious it was, but part of the initial PR about the Grand Challenge included the goal that it would enable the US military to convert some significant fraction of their fleet of ground vehicles to autonomous operation within ten years.
I do think the previous poster was overstating their case a little, but not so badly as you seem to believe.
[0] https://en.wikipedia.org/wiki/Principle_of_charity [1] https://news.ycombinator.com/newsguidelines.html
> In 2006, Volkswagen showed off a self-driving version of its Gold car, named "53 plus 1" in honor of the original #53, Herbie. Autoblog.com posted a video of the car in action.
Should I go further back in history? Because we can.
Yes, the concept is revolutionary. But so is matter teleportation, and that's not going on sale next month either.
What do we call technologies that have been hyped for 10+ years? This is not just a knee-jerk contrarian point of view. Vaporware is actually the correct term we use for these products, not a slur.
It's not bullshit revisionism. People really have been talking about these things forever but not delivering them.
> Why do you feel the need to lie about recent history? Shouldn't you observe that if you have to make a nasty lie to support your argument, maybe your argument is wrong?
Settle down, nobody's doing any nasty lying, see the link above, this discussion has been going on longer than 10 years.
In contrast, fission reactors have been around the corner for a long time, however, there if you use the metric of net power generated from fission as a metric, no real progress is being made there.
I see a real difference between these two "always around the corner" technologies.
https://www.google.com/search?q=fusion+progress&source=lnms&...
I would encourage humility when considering whether to make a claim about a field that you are not well informed of. Real progress is being made, by a number of different groups, in a number of different directions.
But you have touched some use cases that are more frequent than most people think. London is a great example. My hometown Athens is much more closer to London as far as driving is concerned as opposed to LA or NY. I guess many other metropolis also.
How will AI brand A communicate with AI brand B about life critical decisions ? Hey we can either both crash to each other or kill that kid on the pavement. ( just an example , I don't want to go to the morality of the situation)
I assume there is going to be a "standard " . But have a look at your everyday tech. Standards are not easy. There are still WiFi routers/repeaters that cannot talk to each other nicely, just to give an example.
Usually in life critical systems you don't have interoperability issues because there is no interoperability. Huge fail-safe systems from a single vendor.
You don't use your general AI to see someone extending their hand in traffic ex. General AI is the ability to both drive car, score a girl, cook food, clean your room etc.
I think you would be surprised how little general AI have to do with this.
Anecdote: I was driving down a long, narrow country road, lots of turns, space only for 1 car, land rover coming in opposite direction, and another car behind me. We had to verbally negotiate who reverses back 50m to let the others go past. I cannot think how an ML algo could resolve this situation, even if it could, how would it communicate it's intentions to the humans, when flashing your lights isn't enough?
Driverless cars will never (until we have general AI) be able to fully replace a human driver. However we don't need to get to 100% for the technology to become useful. It might be that driverless cars will be restricted to only driving on certain mapped roads. Or every car's movements are tracked in a central control room and if a problem is detected either by the car or the passenger it could be controlled remotely. Or let's say a GPS sensor breaks an actual human driver might be issued to come and rescue you. Lastly we might even accept a certain failure rate. Society is willing to tolerate failure of technology to a certain degree in cars and planes today. All we need is is for the technology to make us all better off on average for it to be successful.
People get this stuff wrong all the time. Especially when they're drunk, tired or texting. These cars won't be perfect but I think pretty good and conservative under uncertainty will be better than human, drunk and overconfident.
The cars don't have to be perfectly safe, just safer than the average driver.
Why are people looking at the extremes? Only human drivers or a completely self driven car? What about a driver assistance system, that corrects a Human's stupid mistakes and still lets the human drive. I'm sure this hybrid (which is the current industry) is safer than both your extremes
Are there statistics?
It's true that humans do stupid things, but so do robots.
What would be _safer_ is an automated "hit" avoid system (if you're about to hit a wall, warn!!) or long distance driving system (for trucks/buses).
I think (and I could be wrong) that there are many more cases where conservative driving is safer than aggressive driving.
In a lot of the cases mentioned, doing nothing or stopping are the safest choices. What's safe to drive over? Just assume everything is unsafe and only drive over it if it puts others in danger to avoid it.
If the road is blocked and you need to communicate to navigate, don't. Wait.
If the road conditions are changed due to construction in ways the car can't understand, stop. Find a new route.
Most humans wouldn't do these things. The nice thing about the autonomous car is that it can drive extremely cautiously.
Traffic lights will send out the same messages they send to humans by red/green lights to the cars using radio.
Emergency vehicles, roadworks, traffic signs etc will do the same.
Or so I assume, because it makes so much of sense. I know there is vehicle-to-vehicle communication in the works: https://en.wikipedia.org/wiki/Vehicle-to-vehicle
I live in the capital city of an EU country, population 2 million, and there's a traffic light about two blocks away from where I live which has not been functional for about 2 weeks now. And the bad thing is that the other traffic lights from that intersection DO work, so you cannot rely on the road-signs, like one usually does when ALL traffic light from an intersection malfunction (from loss of power, mostly). So I'm a little skeptic that traffic lights will be able to "reliably" guide no-driver vehicles using radio signals anytime soon.
Most of the traffic lights in Las Vegas already support vehicle-to-infrastructure (V2I) communication, although you need an Audi A4, Q7 or Allroad manufactured after 1st June 2016 and an Audi Connect Prime subscription ($33.33/month) to make use of it [1].
[1] http://www.theverge.com/2016/12/12/13923254/audi-v2i-las-veg...
I should also add: - very fast mobile data, with no blackspots, so a remote driver can take over a car that's confused by a situation - a replacement for GPS, it's not reliable enough, the signal is far too weak - digital signal beacons embedded in lane markings, and edges of the road, and on traffic cones for roadworks
nope. nope.
The remote takeover seems entirely separate, and both promising and terrifying.
... that's right, now you know what the driving force behind all those flying cars in sci-fi was. It's simply a way to separate the autonomous and old-style vehicles.
I thought it was that winged flight makes for crappy cars and wingless, vectored thrust flight is horribly fuel hungry. (Sci-fi versions are usually wingless and either have magic drive systems or just ignore the fuel issue.)
I would also add; that I will stay consider these vehicles to be unsafe until the manufacturers can prove to be beyond a shadow of doubt that I or the local computer are the ones piloting the vehicle.
When a random person can remotely take control of a vehicle; whether it be the windshield wipers, or full control of the vehicle, then it is not safe to contain a power source in the vehicle near a public road. I don't want 4chan driving me into another vehicle.
If there was some chance my car could get hacked, or other weird vulnerabilities that robocars had that traditional cars did not, I would still want to be in a robocar if my chance of dying was cut in half.
It may be true that the chance for an average person to get targeted multiplied by the chance for a targeted person to get their car hacked is lower than the chance of dying while driving yourself, but that's not reassuring at all if you think you have a higher than normal chance of being targeted or are already being targeted. From experience of being on the receiving end, it's really not that hard to piss off 4chan or get unwanted attention from other crazed parts of the internet. (For an extreme case, consider a certain pizza place in the news lately...)
My only questions are
1) will the skiddie tools be on github first, or will someone have to buy/leak them and repost on github?
2) If a minor executes the scripts and lives are taken; and assuming you can determine who ran them, will they be tried as adults and will the verdict or punitive actions make up for the damages?
It's also not particularly clear that the costs will drop rapidly to zero or even to a level that leads to mass adoption (see air travel at Mach 1+, Segways...).
If I had to compare, I expect the change to be more akin to what Excel has done to paper spreadsheets.
Good sensors aren't cheap, the optical and mechanical elements aren't likely to drop in price soon. Only the computing power is getting cheaper, but harnessing it isn't particularly easy (we're not getting better at programming as quickly as we're making faster chips).
If you need another example, consider automated cleaning robots. It's a considerably simpler problem to solve, and they're still only a marginal product. At the moment, autonomous tech is only really used in well-controlled industrial environments. It will surely break out of there progressively, but chances are it will take a while.
So if all of America's 11 million professional driving jobs were to disappear over the span of a decade, it would be a blip, not an apocalypse.
It's still NHTSA level 3, not 4 or 5, but it's a major step up from pretty much anything currently available and will be in consumer's hands. Even more so when the Model 3 comes.
I think you'll be surprised how fast this technology will exit the labs. Just that Google has decided to brand around it is a big indicator that they're growing confident.
The "goal", such as it is, is to have a technology worth using. It's not to transcribe whichever comic book utopia someone has in mind.
If she had been in an automated car, it would have almost certainly have stopped before the collision.
And what's wrong with slowing/stopping for a plastic bag? How often do plastic bags really sit in the street shaped like babies?
I would shorten that to "other humans", to specifically include pedestrians. That 1/2-second eye contact between driver and human is key. Would anyone here dare walk in front of a car without that little nod from the driver?
The other edge case, one that is never discussed, is how these things expect to deal with off-road driving. I regularly use car ferries. Getting on and off those, and negotiating the terminals, stumps many human drives.
They just have to make them safer than human drivers.
And these days with everyone swerving as they text, the bar isn't exactly set very high.
That system can fix the negatives of human drivers without affecting the advantages of humans over machines
Now you've reduced all your problems down to simply "identify problem and alert a remote human".
If each of those problems only happen once per day, and they are resolved in 30 seconds, it's going to cost only about $30 per car per year to have that remote expert on hand.
I think we'd need a huge infrastructure upgrade to make life easier for autonomous cars. It'd have to be >100 Mb/s mobile data, with extremely good coverage (for the recovery mechanism you mention). Think GPS would need to be replaced too, the signal is way too weak and unreliable. We'd have to embed digital markers/emitters alongside lane markings, traffic lights, signs etc, so they're easier to deal with.
Connectivity you see is from a single mobile phone. A car could have a better antenna, service from all the mobile networks rather than just one, wifi fallback and a backup satellite uplink. If all those fail, you dispatch an engineer to find the car wherever it's stuck and add a solar powered LTE repeater.
If you have a map of the whole service area and know where you started and have a camera looking out the window the whole time, there is no need for GPS really.
Considering these are car companies doing this, true remote control is not a good idea. The industry response to the car hacks was pathetic. Only Tesla reacted properly. (And probably Google...)
Comparatively, the car did not attempt to do any of this and actually didn't even realize its cameras were in a state of failure.
But that's an obvious chicken or egg problem that has unrealistic solutions like making sure everyone starts driving a networked car at all once.
Something I've never seen adressed is how SDV fleets behave when change occurs. Are there chaotic oscillation patterns that emerge from AI trying similar avoiding behavior.
There's research in massive multi agent behavior, but I've never seen it tested on the field with SDV makers actual models.
It doesn't need to be perfect. It just needs to be better than average human drivers. Sure there will be cases it can't handle. But the same is much more true for human.
- communication with other human drivers. In London, this is required all the time, like when parked cars block the road, allowing just one car through. Or traffic light out of action, so you negotiate with other cars using hand gestures, light flashes etc
- endless roadworks, that change what lane you're allowed to go on, turn a two way road to one way road.
- random debris on road. Plastic bag - safe to drive through, wooden plank - safe, plank with nail - not safe.
- loss of GPS, mobile data, or both (again, surprisingly frequent)
- making way for emergency vehicles (sometimes need to drive into lane you're not normally allowed to go, I.e. Bus lane, pavement)
- policeman coordinating traffic
So far, I haven't found any evidence of autonomous cars dealing with the above. If anyone has, please post.