cars that drive themselves.

[dstebbins]

In the Science Channel Documentary 2057, they talk about cars that drive themselves. They say that they're still working out the kinks, but I find that to be hard to believe, mostly because I've seen (and even operated on a small scale) this technology first-hand.

 

It's called an AGV (Automated Guided Vehicle). When I was in high school, I took a class called Technology Discovery, and one unit was on the AGV. Its course and work schedule are literally as easy to edit as the contents of a spreadhseet. If it finds a barrier, it will take a detour. Also, you can program it to stay on a dark black line (a "road" of sorts). The only problem is that this technology is so expensive that only companies can use it.

 

For the other half of the equation, we have GPS navigation technology. This basically tells you what route to go to get to your destination, but you are responsible for driving there. If you make a wrong turn, the GPS navigation system will recalculate your course in a matter of seconds. Granted, they are still working out all the kinks, but it's nowhere near the price of an AGV, so I guess you could say that you're paying for less bugs when it comes to an AGV. Besides, the Tom Tom brand of GPS is actually quite reliable as far as staying up-to-date is concerned, so I've heard.

 

Did it ever occur to anyone that we could just combine these two technologies, and make a smart car? The GPS technology automatically programs the AGV car, and the AGV car goes where it is told, and informs the GPS of any obstacles it encounters, so the GPS can rework the navigation, and the cycle continues.

 

I honestly don't see why guys like Donald Trump and Bill Gates (ESPECIALLY Gates, since this is a matter of COMPUTERS) haven't already gotten these smart cars, because for the life of me, aside from the cost, I cannot figure out a problem with it.

[insane_alien]

the problem is that there is not just one car out there. and they do use both of those technologies and more. machine vision is the main one as it lets the car knowabout the road even if it isn't a 'smart road' (ie one that makes the job of the car navigation system easier), other hazards and other road users.

 

i'm pretty sure if you got a few thousand of those smaller automated cars, chucked them on a model of a road system and let them run free all going to different destinations you'll soon result in chaos, traffic jams and collisions. If it WERE as simple as you seem to think it is then we would have it already.

[bascule]

There's been a lot of people working on this for a long time. It's been a tough nut to crack.

 

Most (if not all) of the vehicles in the DARPA Grand Challenge use a synthesis of many different tracking and navigation systems in order to drive a car. This most certainly includes GPS.

 

The DARPA Grand Challenge has now shifted from driving vehicles in an otherwise empty desert environment to an urban environment. We'll see how the DARPA Urban Challenge goes.

 

There's lots of information available on autonomous vehicles here:

 

http://en.wikipedia.org/wiki/Driverless_car

[Mokele]

Part of the problem is speed and size.

 

Most animals have downright crappy vision, and seem to simply bumble along (beetles, for instance) letting reflexes and mechanical properties sort out all but the most severe perturbations.

 

But cars are very big things moving very fast, both of which dramatically increase the level of damage suffered by even a minor collision. The speed also means the operator (computer or human) needs detailed information not just about the current environment, but about the environment several hundred feet ahead, and the ability to rapidly anticipate and predict the movements of other actors (vehicles, pedestrians, falling rocks, etc.).

 

Basically, the only reason cars work at all is because the operator has a HUGE visual cortex and some of the most advanced vision in the animal kingdom (seriously, humans and other primates are up there with hawks, compared to most species). Had we evolved from wolves or possums or lizards, we'd probably consider much slower speeds to be the limit of "safe".

 

However, I think there's still hope. We just need to 'think outside the box'. We could try a system like jumping spiders, with two eyes permanently fixed on long distances, and 6 others set to examine the nearby area. We could implement electromagnetic sensors all around the car, like the lateral line of fish. We could even add super-elastic bumpers to absorb the consequences of errors.

 

IMHO, the bigger problem will be public acceptance.

[insane_alien]

another downside is that your fuel consumption will increase due to the power requirements of the computers needed for machine vision.

[bascule]

However, I think there's still hope. We just need to 'think outside the box'.

 

The nice thing about having a computer operate a vehicle is that you don't have to depend on evolution to determine its "senses". You can use whatever ones make sense.

 

So, think: cameras, RADAR, lasers, GPS, sonar, whatever, all working together to help the computer figure out where it is and what's going on around it.

[insane_alien]

So, think: cameras, RADAR, lasers, GPS, sonar, whatever, all working together to help the computer figure out where it is and what's going on around it.

 

and the cars can have transponders the inform other cars of its intentions so they can plot actions to result in minimal disturbance.

[padren]

To go a bit further on the visual point... imagine a piece of cardboard on the road. A human can quickly evaluate it and determine it is safer to drive over it than swerve around it - especially in moderate traffic. That cardboard could be half folded sitting on it's side - and look like a 3D large obstruction. It could be flat but still take a lot of data to process it's flat due to an odd shape. It could also be blowing across the road tumbling, and be visually interpreted as an animal or person - and critical to avoid.

 

These things are easy for humans because not only are we good at processing visual data but we have usually processed just about everything we are liable to experience on the road for 16+ years before we even start driving. That's a long time to practice visual identification.

 

 

And second point (at least in reference to Bill Gates or others researching it) that is important:

 

Liability:

It would have to be very demonstratively safe. The thing is even under ideal situations humans with exceptional driving ability do get into accidents, and at times are considered at fault for those accidents. What happens when a car does that due to:

 

1) a software flaw?

2) a "could have been better" algorithm?

3) a choice that was wrong but seemed right and would normally be right? (Sometimes you have to take the 60 in a 60/40 choice, but you'll still be wrong ~40% of the time)

4) a combination of the above and vehicle wear?

 

A guidance system vendor couldn't guarantee perfect driving, but it could be hard to nail down the exact amount of risk the buyer is getting, with contentions on all sides when something goes wrong.

Edited April 8, 2009 by padren

[insane_alien]

At the very least the car would have enough telemetry stored to determine exactly how a traffic accident happened and what was to blame. so it would reduce false convictions at any rate.