Quote: Original post by jtagge75
Except when it makes an error tracking a human you just have to restart the level. When it makes a error tracking a cat running across the street your car runs into a wall. I'm not quite as willing to place my life in the hands of machines until the technology has matured more.

The thing is the problems tracking humans have to do with tracking pieces of humans. That is very complex. For the purpose of a car, the car just needs to know there is something there, which is pretty simple with kinect like devices. All the car cares about is that there is a blob of stuff there that it doesn't want to hit and whether that blob of stuff is moving or static.

The problem with kinect in that regard is that in the open there's a lot of infrared interference.

Oh, yeah, and there already exists technology along this line in some cars.

Top Gear has shown a car which can more or less park itself (when given a location by the human operatior), a car which can drive their track perfectly once shown how and a car which automatically tracks the speed of the car in front and speeds up/slows down as the car in front does without requiring human input.

Most of these systems however are on expensive or experimental cars.

Quote: Original post by phantom
Top Gear has shown a car which can more or less park itself (when given a location by the human operatior), a car which can drive their track perfectly once shown how and a car which automatically tracks the speed of the car in front and speeds up/slows down as the car in front does without requiring human input.

The ones you describe actually aren't on experimental cars. I think one of the lexus' has it standard, and it's really ntohing to write home about. It just does the same thing you learn in driving school and has a sensor that just says when anything is within 2 feet or so of the rear bumper/front bumper, not where it is in space.

Far more interesting are the cars that participate in the DARPA grand/urban challenges. They can drive completely unassisted in traffic.

Quote: Original post by way2lazy2care

Quote: Original post by jtagge75
Except when it makes an error tracking a human you just have to restart the level. When it makes a error tracking a cat running across the street your car runs into a wall. I'm not quite as willing to place my life in the hands of machines until the technology has matured more.

The thing is the problems tracking humans have to do with tracking pieces of humans. That is very complex. For the purpose of a car, the car just needs to know there is something there, which is pretty simple with kinect like devices. All the car cares about is that there is a blob of stuff there that it doesn't want to hit and whether that blob of stuff is moving or static.

The problem with kinect in that regard is that in the open there's a lot of infrared interference.

I didn't think about the infrared interference but this gave me an idea for something kinda cool. What if you had a car with a robotic arm on it's side right near the gas cap and gave it the ability to automatically drive it's self. The car also has a basic AI. Basically it could call news paper places and ask for a job delivering papers. To communicate it would just use something like this: A. L. I. C. E. with voice recognition and text to speech. It could explain that it needs the papers taken out to it and it's willing to work for practically nothing and it's willing to take any hours it's given.

Most places would hang up the phone hearing a robot voice but since it's a car it can call and drive all over looking for a job. It would only need Gas, Insurance, AAA, and a cell phone plan. If something happens and it gets a flat tire or something it could call AAA and ask for a tow to the nearest auto shop. It could pay for it with the money it saves up from running paper, delivering mail, and maybe even pizzas.

It's just a crazy thought but I could totally see it being done with todays technology.

Quote: Original post by SteveDeFactoI personally think it would have been better used in the auto industry. Imagine cars that drive for you and almost never make a mistake. It could actually be saving liefs instead of just being a cool toy.

The auto industry and researchers have been working on this stuff for years: pedestrian tracking, line tracking, sign reading, tracking driver eye movement etc. I assumed the Eye Toy and Kindle was based on this and other related research and were simply mass market implementations.

There are also robot cars that can do a fairly good job of driving themselves in sparsely populated areas, but they never let them drive autonomously in built-up populated areas for obvious reasons.

Quote: Original post by Trapper Zoid
There are also robot cars that can do a fairly good job of driving themselves in sparsely populated areas, but they never let them drive autonomously in built-up populated areas for obvious reasons.

they let the car that won the DARPA urban challenge drive in downtown san francisco. There was a guy sitting in the front seat with his foot on the brake ready to take over, but it didn't crash into anything :-p.

The final for the urban challenge actually had a decent amount of cars on the road. Not rush hour traffic, but easily middle of the day traffic for any moderately populated suburb shopping area.

Quote: Original post by phantom
Oh, yeah, and there already exists technology along this line in some cars.

Top Gear has shown a car which can more or less park itself (when given a location by the human operatior), a car which can drive their track perfectly once shown how and a car which automatically tracks the speed of the car in front and speeds up/slows down as the car in front does without requiring human input.

Most of these systems however are on expensive or experimental cars.

I'm amazed so few people are aware, but Google have already done this in real life, with cars that drive themselves on real roads.

Link

Quote: Engineers at Google have tested a self-driving car on the streets of California, the company has announced.

The cars use video cameras mounted on the roof, radar sensors and a laser range finder to "see" other traffic, software engineer Sebastian Thrun said.

They remain manned at all times by a trained driver ready to take control as well as by a software expert.

Google hopes the cars can eventually help reduce road traffic and cut the number of accidents.

In a posting on the company's official blog, Mr Thrun said the self-driven cars had so far covered 140,000 miles on the road.

They have crossed San Francisco's iconic Golden Gate bridge, negotiated the city's famous sloping streets, driven between Google offices, and made it around Lake Tahoe in one piece.

I laughed a little at the notion that something like Kinect would be the base of new technology used in other fields. I'm quite ignorant when it comes to the development and progress of modern technology, but it seems obvious even to me that the technology used by Kinect has existed for at least five years and performs far better today (for what it's intended to do).

Since Kinect is a mass market product, its technology may be new to that market, but relative to modern technology I'd guess it's both behind and inferior.

Self-driving cars have been available technology for over a decade. Nobody wants to develop them into products, because customers would rather take a 10% human-administered failure rate than a 2% machine-administered failure rate.

Quote: Original post by superpig
Self-driving cars have been available technology for over a decade. Nobody wants to develop them into products, because customers would rather take a 10% human-administered failure rate than a 2% machine-administered failure rate.

I remember documentaries about them in the early '90s. There have been many different editions of the technology ranging from embedded tracking devices in the road to onboard cameras and tracking devices.

Today's automated vehicles are amazing. The annual DARPA challenges are cool and various studies and documentaries show great progress. Automated vehicles can certainly handle most real-world driving conditions.


The big problem is still what to do under exceptional circumstances.

First, the machines often have difficulty coping with certain mechanical failures and with certain external road conditions. A surprisingly high number of DARPA challenge vehicles fail for things like tumbleweeds, wind-suspended debris, and other destructible objects in the road. Others fail with a bit of dust or rain or snow in the air. Sure, they are great at exceptional circumstances such as watching another out-of-control vehicle or responding to emergency stops, but they still have a lot of situations to work out.

Second, there is the whole issue of legal accountability under exceptional circumstances. It is one thing to bring a human to court and deciding if they acted rationally under the circumstances. The damages are realistically capped when an individual is sued. It is something else altogether when the same people take a megacorp to court and attempt to prove that their machine took appropriate steps in the circumstances. Not only is it (currently) difficult to defend the machine's actions, but many people and lawyers get greedy when the defendant is a corporation with deep pockets rather than an individual.



I'm with you that I think automated vehicles are a better alternative and are nearly ready for mainstream use. I just don't think society is ready for it.