Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[scrdest]

There will have to be some *VERY* aggressive security and digital signature enforcement for autodriving cars using GPS and short distance vehicle-vehicle data transmissions, before I would even THINK of entering one.

A malicious map update with a Man-in-the-middle scenario can easily make a car think the fastest route is through the lake, of off that unfinished bypass.

Vehicle-vehicle communication intended to keep cars from hitting each other can be abused by hackers compromising the car's computer control system, so that they can force people to brake for them when the douchebag in traffic, or even initiate an accident from the roadside with false messages causing cars to suddenly stop to avoid a car that isn't really there.

All kinds of things.

The potential for abuse in an AI driven vehicle is astounding.

You of course, will also have the black market modification scene, for people that just can't feel alive without doing 50mph over the posted speed limit, but don't want their insurance company to know, and or-- want all those dumb robot cars to get out of the way for them. (Fake having a police of emergency vehicle identification signal so people pull over, et al.)

Without some really crazy strong and properly implemented security on the computer to prevent tampering and unsigned data from being used, automatically driving cars are a blackhat's wet dream, and a very lovely means of performing a political assasination.

Which is why nobody important would use them. No big deal, if you are powerful enough to be worth assassinating, you can afford a driver.

And regular people? Why bother, and tampering with pretty much any systems could lead even accidentally to the person dying, which means Urist McHacker is now guilty of manslaughter which is not really a thing people are willing to let you get away with easily. It would be like trying to stand on the roadside with a rifle and trying to shoot people's hood ornaments off - not really that bad, but you hit someone or something important and you are thoroughly screwed.

[kaijyuu]

Because they are not widespread yet. And also hackers that could potentially do everything in the road are not nice.

Current cars have the exact same vulnerability.

Charlie Miller, a security engineer at Twitter, and Chris Valasek, director of security intelligence at security firm IOActive, aimed to increase awareness of car hackability by hooking up a Nintendo game-console controller to a US-market Ford Escape SUV.

They were able to accelerate, brake and steer as though they were playing a video game. Except this wasn't a game. It was a very real two-tonne SUV and it had been comprehensively hacked. Miller and Valasek also wired into a Toyota Prius hybrid car using a laptop computer and took control of several safety-critical systems including the brakes.

Making a device roughly similar to a garage door opener that disables a car's brakes can be done today.

[Bauglir]

I'm honestly more worried about the sheer technical aspects of it, and less so about making Hollywood-style hackers a little more plausible. How's the vehicle-to-vehicle technology hold up with cars going 70 mph in opposite directions on an interstate as part of packs of a few dozen vehicles each in shitty weather? If it's anything more complicated than just "I am here" beacons, I'm skeptical of how well it's going to work. If we can get networks communicating that fast and that reliably in conditions that adverse, I'd like to call bullshit on the router industry for making it take so long, and be subject to so much interference, to connect to a wireless router in a crowded apartment building.

[kaijyuu]

cars going 70 mph ... in shitty weather?

Luckily this won't happen because AI controlled cars, unlike humans, aren't fucking stupid.

[wierd]

I'm honestly more worried about the sheer technical aspects of it, and less so about making Hollywood-style hackers a little more plausible. How's the vehicle-to-vehicle technology hold up with cars going 70 mph in opposite directions on an interstate as part of packs of a few dozen vehicles each in shitty weather? If it's anything more complicated than just "I am here" beacons, I'm skeptical of how well it's going to work. If we can get networks communicating that fast and that reliably in conditions that adverse, I'd like to call bullshit on the router industry for making it take so long, and be subject to so much interference, to connect to a wireless router in a crowded apartment building.

The issue with wifi routers is that they have to use a frequency band that also happens to be the absorption band of water vapor.

That's the reason why microwave ovens use the exact same frequency! (2.4 to 2.8ghz)

Because of this the signal becomes jumbled energetic noise very quickly. Perversely, that is *precisely* why this is the unlicensed consumer band. Anything with any real carring capacity is too valuable to allow the public to just use-- why, that would be oh so unfair to the cellphone and tv broadcasting industries! Something that valuable MUST have a licensing fee attached!

Sarcasm aside, because a wifi router has to cope with its signal being turned into unintelligable crap just from the humidty in the air, it has to emit a fairly sizable amount of energy to have a reasonable SNR over any reasonable distance. The reason why walls block the crap out of wifi signals? The whitestuff in drywall is hydrated gypsum. Calcium sulphate hepta hydrate, IIRC. Eg-- "lots and lots of chemically bound water".

2.4ghz is possibly the worst possible frequency to broadcast data over. That's why it has problems with being crap from the start.

A vehicle -vehicle communication system will have to be on something closer to 300 to 600mhz, and will need spectrum licenses. Those frequences have good fidelity at low broadcast energies, and don't attenuate nearly as quickly as 2.4ghz.

[alway]

Fortunately, we're a long way off from having the reliability necessary to exceed even human safety rates

There is approximately 1 accident per 300,000 miles driven in the US*

Google's self driving cars have gone well further than that (around half a million miles, last I heard) without accident. Which is to say, barring a massive swath of several accidents, it already surpasses human safety rates, and has for a few years.

Thing is, we are terrible at driving. It's a task which is both highly complex and incredibly dull. People get bored, distracted, drunk, or sleepy. Self-driving cars never do. If you had a self-driving car, your insurance costs would drop enormously, as would your risk of being one of the over 30,000 people who die each year from wrecks in the US alone. It's similar to the number of gun deaths in the US, and that's only if you include both homicides and suicides. If you can even cut that number in half, it's the equivalent of stopping a 9/11 every other month. Driving is really dangerous compared to just about everything else we do. Of course, it won't just cut it in half; it will cut it nearly to 0; which is precisely why self-driving cars are being pushed so hard.

Now, if you DO want to learn about how they work, I highly recommend looking up info about them Sebastian Thrun has put out (he was the guy who headed the Stanford self-driving car project, then the Google self-driving car project). The most interesting stuff, though, is visualized debug information: http://www.youtube.com/watch?v=YXylqtEQ0tk (Also, at 12:00, you get to see another driver freak out seen from the perspective of their laser range data) (At 14 minutes, you can see it doing a reverse drifting parallel-park). The more you learn about them, the more safe I suspect you will find them to be.

*Approximately 10 million accidents per year; approximately 3 trillion miles driven per year

[Bauglir]

Pre-Edit: This section rendered largely irrelevant, but this post would be pretty barren without it.

Ehh. Maybe I'm overly cynical, but it seems like if licenses were going to be made available to the general public in the numbers required to support a hundreds-of-millions-of-cars environment, they'd already have been made available for other applications. Unless this is the way in which we're going to incentivize reducing pollution by cutting down on vehicles on the road, there's going to be a huge number of the transmitters operating in one area at any given time, and you won't have any reliable way of telling in advance which frequencies in the available bands each one is operating on.

Also, it wouldn't take many accidents to get the whole technology to go nuclear in public opinion - that is, actual efficacy be damned, it's "too scary". People don't like to feel like they're giving up control, they don't like change, and they think they're better than average drivers that don't need it. So I'd expect people to be just waiting for something to go wrong so it can be decried, even if the actual statistics demonstrate an improved safety rate. This isn't an objection so much as a "We'd better make sure it's goddamn perfect before it goes online" comment, though.

NINJA: Well. Nevermind, then.

[MetalSlimeHunt]

And mind you, the decline in crash rates that we have observed is only for individually self-driving cars. They may well become even safer when there are enough on the road that they can communicate.

[wierd]

The issue I have isn't that I don't trust the car and its sensors to do unexpected things.

It's that I trust it will do exactly what is expected of it, when it follows its instructions.

I program robots for a living. Not nearly as sophisticated as a self drivng car, just CNC machines, but I know full well that if I tell the robot to do something stupid, it will do something stupid, without fail.

As long as there is a means for an untrusted person to inject instructions, the very reliability of the robot is what makes it dangerous.

I want at least 4096bit elipitcal curve crypto (without the NSA's crappy number generator) behind the digital signature security for data updates, and I want protected hypervisors monitoring the conrol system's process space.

[Bauglir]

And mind you, the decline in crash rates that we have observed is only for individually self-driving cars. They may well become even safer when there are enough on the road that they can communicate.

Considering I thought it'd be harder to make it work without communication, I'm pretty well-convinced at this point.

Maybe I'll just install one of those toy steering wheels so I can still feel like I'm driving >______________>

[wierd]

Sadly, when 99.9% of all traffic is fully automated, then programmers will become lax. (Because 99.9% of traffic is automated, and predicting stupid human behavior is hard yo.)

When the technology is very mature, it will have to be illegal to drive without the autodrive and avoidance systems on.

Otherwise, jackholes that need to do 50mph over the speed limit and drive with their heads up their bums will cause all kinds of problems, and just be a suffciently edge case as to be relevent only as media fodder. (Like plane crashes. The odds of being in one are very low. But when they happen, they happen big.)

If you make the porche obey the speed limit like a good little robot, you deny the machismo governed tools the "rush" the vehicle represents. Might as well buy a geo metro instead.

Expect black market mods.
I'd bet money on it. You already find updated fuel mix ratio eeproms for the ECS systems in existing cars for these tools on the open market.

[alway]

And mind you, the decline in crash rates that we have observed is only for individually self-driving cars. They may well become even safer when there are enough on the road that they can communicate.

Actually, I doubt that will make them safer by more than a token amount. They're already good enough at navigating that direct communication isn't really necessary. When you've got a 360 view of your surroundings, a relatively large object like another car isn't very likely to be hit.

The big benefit will be traffic routing. The reason most traffic jams form is because of imperfect routing combined with imperfect reaction time.

Imagine you're stopped at a stop light; the light turns green. Since you are behind 12 cars, you sit another 15-30 seconds before you are able to hit the gas pedal and move, since it takes a while for the entire line of cars to start moving. The reason for this is twofold: reaction time and safe following distance. You wait a second or so because of reaction time; then another second or two for the car to give you a safe following distance. Meanwhile, more cars are slowing down and stopping behind you, despite the light being green, because you are blocking their path. If you know all the cars ahead of you are able to do so as well, all 12 cars can communicate with one another, and all agree to start moving at the same acceleration as soon as the light turns green. Bam. You just cut the city's congestion problem in half, vastly reducing commute times, pollution, and putting gas money back in people's pockets. It's a natural outgrowth of self-driving cars, along the lines of earlier thoughts: http://en.wikipedia.org/wiki/Platoon_(automobile)

Likewise, cars could communicate traffic conditions; not only with themselves, but potentially with some external system. Highway backed up and going 5 mph? Automatically take an alternate route. To some extent, todays' GPS systems do this already (if I had a Bluetooth connection with my phone, my car's GPS could do so today), but it would give a very up to date look if given information by cars going in the opposite direction on the highway which happened to see your lanes slow down.

[Tsuchigumo550]

I'd get a self-driving car if I didn't live in Fucking, Nowhere.

I'm transhumanist, but... idealistically so. I'd love a longer life, more direct control over what my body wants and does, that sort of thing, hell, I wouldn't mind something as stupid crazy as reinforcing my bones or rearranging my organs for better survivability, but I'd like the process of having it done be safe.

I'd be totally cool with replacing my legs, actually. Knees especially. I kind-of-sort-of have a "better foot" design concept I've always wanted to try (basically, you stand on a ball with six "toes" coming out in a hexagon shape, the "toes" are the main source of traction and can have various amounts of tension applied for climbing or just walking) that I came up with because I draw mecha all the damn time. We already have some crazy awesome prosthetics, and while this makes me sound like a body modder (i'm not... that much. Most I've ever actually considered this day and age is a tattoo, and not a prominent one either) I want mecha legs.

One of my knees is shit and probably always will be, as there's no real need or ability to undergo surgery, but it still acts up and moves way too much. I'd like to keep my fun bits, but aside from that, the waist down can go elsewhere. Unfortunately, there's a lot more human in the legs than just legs, and it's gotta go somewhere, and all those veins and the ilk aren't going to survive removal... so there's issues. Issues that I'd like to tackle.

Of all things I plan to be a game designer, but I've put serious thought to engineering and medical practices before (lack of drive is what stops me, mostly.) and the marriage of the two is something that I love the idea of and realize makes other people think I'm psychotic.

And maybe I am. I'm the guy who's talking about getting mechanized legs for no good reason other than a slightly off knee that still works fine. Let's say I woke up tomorrow with these things. I'd trip every metal detector from here to Kansas, have a very difficult time driving a normal vehicle, and my god, the logistics of it all; I'd be practically landlocked due to weight, and then there's upkeep...

So, maybe I'd want interchangeable legs. One's a carbon-fiber based model made to simulate the human leg, and then the model I envision is like... the hummer. Stupidly excessive in the best ways.

Something feasible that I want in my lifetime? Control switches. Link my stomach and my brain so that I can stop feeling hungry when I've clearly eaten enough. Imagine, even if I had to manually input data on the meal I just had/am having, the computers inside me would accurately tell my mind to sense fullness when I've got enough calories out of my meal. Perfect. I could set it to dieting, or for a period of high or low activity.

-------

On the cars...

What if you had a failsafe? Like, I realize, hackers could probably try to disable it somehow, but have myriad systems that start screaming "Something is wrong!" and transmit that like crazy to whatever's nearby. A crash happens. Other cars read the "Something is wrong!" signal, and can react? In addition to sensors?

[MetalSlimeHunt]

When the technology is very mature, it will have to be illegal to drive without the autodrive and avoidance systems on.

Otherwise, jackholes that need to do 50mph over the speed limit and drive with their heads up their bums will cause all kinds of problems, and just be a suffciently edge case as to be relevent only as media fodder.

If the technology is mature, the machines will be the ones who are 2fast4you. Remember, computers have faster cognition than humans. They have no reason not to be going 2fast2furious all the time, not even remaining human drivers (who they can simply avoid at 130 MPH).

[WillowLuman]

I seriously doubt cars will become unable to be driven manually - that would require every road in the country to be digitized. As offroading and scenic drives are why people with money often buy cars, the manufacturers will probably cater to their demands.