artificial intelligence program to drive a car

the AI driving program is long and complex but it does what it suppose to do which is drive a car safely from one destination to the next and avoid cars on the road and stop when there is a red light. the real program is written in c++ and is a fairly large program. I wanted to scale down the program and convert it to java so i can post it online. here is the link:

http://www.humanlevelartificialintelligence.com/demoinstruct2.html

if you have any comments, criticisms please post it on this thread.

the idea of an AI car is to put it in a real world situation where the AI car can drive a person from one place to the next in the safest and fastest way possible. It will follow all traffic laws including speed limit, traffic lights, road lines, conficting rules, and any obstacles that may harm the car or the person in it. "zero" car accidents is the key here. The AI will also have the ability to learn from past experiences. This is important because iif driving laws are changed the AI car can adapt.

 

Log in or Sign up to hide all adverts.

The stuff you talk about on that page sounds a bit like reinforcement learning. I couldn't get your java applet to do anything though.

 

Log in or Sign up to hide all adverts.

the applet is quite large for a java file: 800 kb. for me it takes about 1 minuite on a 56k modem to load on the browser. You also need a java runtime which is a plug-in.

reinforcement learning, that is exactly what my program is. I took shannons chess program (built 50 years ago) and modified it to work in a universal setting. In his program he only applied it to the game of chess. In my program i wanted to universalize everything -- the enviornment, actions, predictions -- so that it can not only play one type of game but "all" games.

I also solved his problem of infinite possibilities by including something called the deviation functions. The pathways are built so that the possiblities of future actions are universalized.

 

Log in or Sign up to hide all adverts.

ever heard of: Intuitive Parking Assist/ Advanced Parking Guidance System (APGS)

 

"They did that on "NCIS" last week. ^_^ It was"

hahahaha. its just a show.--isn't that the show about that nerd chick who built the AI car and she got gased in it? the real thing won't be hackable. failsafe is the first priority and the technology won't be sold in the market any time soon because it must run through a lot of tests.

but imagine if we can live in a world where "zero" car accidents. look at the car drivers today, we got 100's of millions of car accidents each year.

This stuff can be tested--we let it drive a car in a videogame (playstation 3)and run it for years and years(24/7). if the car gets into one car accident that means it failed. if it didn't that means it works.

 

I cause a car accident and I don't remember. Same case here

Please Register or Log in to view the hidden image!

 

note: for those technical people out there. the AI uses a self-learning "network" that focuses on patterns rather than what is visually there. This isn't an expert system that people have to manually input in order to understand the rules of driving, it learns all by itself by observing how expert drivers drive. Conflicting rules, rules of driving, changing of those rules (all the rules) will be solved by the patterns.

in other words i don't have to input manually how to behave when there is a traffic light, or that the car has to drive in the center of the road, or that when there is a pedestrian stop the car. it learns all by itself -- a universal type of artificial intelligence.

 

a new version of the AI car videogame has been posted on my site.

http://www.humanlevelartificialintelligence.com/city/demo5.html
http://www.humanlevelartificialintelligence.com/aifolder/demo4.html

this one is smarter and faster than the previous demos. it updates itself every second and plan out future actions. it also has fuzzy logic functions to distinguish one image layer from the next. the training required for this car game is less than a minuite to understand the rules: go around cars, stop on the street light, go when its a green light. in old neural networks the training takes weeks and months.

 

I think it sounds cool. As long as you could take control of the car without exception. But to be honest, I am looking forward to flying cars more than any other improvment.

 

n0ise and fuel c0nsumption are the biggest obstacles here...

 

"I think it sounds cool. As long as you could take control of the car without exception. "

thats one feature that can be included, but would you drive the car knowing that you can make mistakes or let an expert AI drive the car for you where it will never get into an accident?

 

Yes..., the world is in an energy crisis. And its not just that oil is running out. Its that you keep running out of and get charged alot for one of the most common things is this universe. This reeks with nuty manipulation. At one point oil worked out okay. But we have just about outgrown it.

Contrary to popular opinion, electricity and power are not hard to create. We are drowing in concentrated energy. We have never understood how to put it to work for us on masive scales but that is not much of a challenge for us. We need only connect the dots now that we have a need and recognition that it can be done.

After that I will be able to fly my car around all I want.

 

Thats just it. It is not perfict. A car's AI may one day be the safest way to get around. If you think that people can make mistakes, then realise who made the car and it's AI. It's a person that can makes mistakes. Perhaps from something as simple as a bird or rock hitting a sensor, all that perfection can come crashing down. Or from a wire that brakes. The person inside should have an override to account for any such error and not have to pay for it with his or her life. But if the AI is good, we can let it drive us around until then.

 

Here are very good videos on the AI driven car; I was stunned watching them.

http://www.torrentz.com/3629b99921db0925dc30c3b744089283eaee8e1e

or try this link http://fenopy.com/torrent/Human_v2_0_and_Robot_Race_XviD_AC3_Samples_avi/Mzk0OTY1/index.html

 

"Thats just it. It is not perfict. A car's AI may one day be the safest way to get around. If you think that people can make mistakes, then realise who made the car and it's AI."

would you rather have an AI car that has a probability of 1/1000000000 error or would you rather have a human drive a car with 1/1000 error?

look millions of people get into car accidnets each year. 10s and thousands of people die each year from drunk drivers and careless drivers. if there is a car that has 1 car accidnet in 100 years then which do you prefer on the road?

 

You computer guys are at it again, slowly but surely taking the humanity out of everything

 

As long as there's a manual override (that the computer can't override) then it sounds great.

 

"You computer guys are at it again, slowly but surely taking the humanity out of everything"--cha

purpose of the human race is to gather knowledge. eventuallly they will explore artificial intelligence.

"As long as there's a manual override (that the computer can't override) then it sounds great."--zen

the good thing about the universal artificial intelligence program is that we can control what the AI learns. it can't make decisions on its own. i basically given it one function: follow the best future pathway. The more times you train the network the stronger those pathways will become. since we train it to drive perfectly then it will drive perfectly. the manual override can be used here.

The human level AI is the complete opposite. its a machine that learns on its own and it makes its own decisions. i'm using the UAI to build the HLAI. those car demos i have posted on my site are just simple versions of what i'm trying to show people. hopefully i can post better demos in the future.

 

You should participate in this competition:

http://hardware.slashdot.org/hardware/07/03/22/2234257.shtml