Software > Software

Artificial Intelligence

(1/6) > >>

nanob0t:
Hello, I'm new to the forums.

I am incredibly intersted in the development of Artificial Intelligence.  Construction of normal robots comes easy to me, along with the programming.  Right now, I'm fighting with the development of my ideas.  I'm pretty sure I have all of Artificial Intelligence broken down.

I am planning to construct an animal-like robot.  It will have a variety of sensors and such that allow it to immitate a rodent animal.  The sensors will run through their own processor in order to interpret everything that come to it, and that information will be fed to a central processor.  If power is an issue, I can hook up several sensors and such to one processor, which it will probably come down to.  A central processor will interpret everything that inputs from sensors and such.  After interpreting it will make decisions.

The robot will be able to make decisions as soon as it's turned on.  It will have a natural curiosity to roam.  This is extremely extremly vague.  I have a full document on ideas on my PDA. 


I was just wondering if anyone else had interests or ideas they would like to discuss.  The robot is there.  I am wanting to converse about the programming needed to obtain a level of "intelligence".    This is open to anything.

Somchaya:
An animal-like robot would be extremely interesting. If you could, give some details on the behaviors you want to implement? Like the list of ideas in your PDA that you mentioned?

I'm also interested in such robots, but more towards flocking behavior and coordinated tasks. Still, your project sounds extremely interesting to me so I'd love to hear more details about it.

JonHylands:

I'm doing something similar:

http://www.bioloid.info/tiki/tiki-index.php?page=MicroRaptor

- Jon

nanob0t:
Well, since I see some people are interested, I'll explain.

First off, I'm not going for the robot portion of it.  I'm going for the implementaion of Artificial Intelligence.  The construction of the robot is least of my worries.

Anyway.  The robot is preprogrammed with as little as possible.  I call them instincts.  It monitors its batter power, touch sensors (emulating skin), etc.  When the robot is turned on, I intend for it to start in a state of homeostasis or all of its parts are at equilibrium.  It has natural curiosity so it roams.  The roaming portion is taken as moving to the closest object of interst.   At the object, it will determine if the object is at all usefull by using its outputs on it.   It is constantly logging this, which is a reason I intend for it to have an onboard computer, or perhaps connection via Bluetooth.

If something triggers something, or puts the robot at risk of extinction, it will do something.  To break it down, here's my intended logic process.

-Stimulus-
What do I do about that?
-Decide if there is a positive/negative connotation-
What will I do about it? <<<<
-Perform Action-                  ^
-Log- >>>>>>Failed>>>>>>
-Continue-

The robot copes with a stimulus, and as a result of the logging... It learns.  The positive/negative connotation is a sense of the stimulus being beneficial or not towards the robot's existence.   It is able to date back to these logs to determine wither it should use something again or not.  By repeating this, it will create processes for itself.


I want to construct this, so you can put it in the middle of a field with a power source nearby, and it will start from nothing to being able to explore and create its own environment.   It will be very... animal-like.  Thus how I based the robot. 



Thoughts?
Ideas?
Criticism?

Somchaya:
Yeap, it sounds very interesting. My suggestion: code the instincts and all that in simulation first, with a simulated agent and world, so that you can actually see what the robot "should" be doing, without worrying about how to make the robot do what it should.

Also, the logging from experience and learning from it is great, and doing it to make sure it works in simulation first would be extremely helpful I think. Stuff like determining positive/negative connotations, and how to choose among various things of interest.

Like for example if there's a ball, and the robot walks to it and pushes it, but it doesn't really affect the robot's existence directly, would that be positive or negative?

And, what if the world was set up so that a chain of events have to occur before something good happens? Like the robot has to push the ball, then roll over a button, before the energy station (the place it gets battery juice) switches on. How would the robot figure that out?

These would be the kind of questions you would have to solve, if you haven't already, and probably test in simulation of some sort before running it on the robot.

Just a suggestion, hope it helps! :)

Navigation

[0] Message Index

[#] Next page