Hello!
I'm looking for AI algorithms for strategically placement of objects in some continously changing environment.
Does anyone know of one?

More specifically:
I want to place sensors (radars) in an area of interest and continously track objects although the sensors can't cover the whole area simultaneously.

This is part of a final year project in my university, and the point is to compare "traditional" algorithms with my own solution.

I'm still not clear on what needs to be accomplished. One may place the radars... anywhere? Are there any restrictions? What is the objective? To maximize the area covered by the radar?

-Predictor
http://will.dwinnell.com

Hello and thanks for answearing!

To elaborate:
I have an area of interest in which some sensors (say 10) is to be deployed in a way that they cover the largest possible part of this area.
The thing is that I want to continously keep track of the known objects which after a while enters the area, if say for example an object enters the area, the sensors should continue to track this object even though it's moving into a (by the sensors) non-covered part of my area of interest, e.g. the sensors should be rearanged in some intelligent way.

The first priority should be tracking known objects and the next priority to keep coverage over the rest of the area of interest.

...
For simplicity the sensors are assumed to be perfect (gives the tracked objects exact location).

Data fusion for sensor data can be assumed implicitly.

The measurement is how many of the objects the system detects and continously tracks. This is what I want to compare.
...

What about deploying the radar units spread out to cover as wide an area as possible, and move them to "chase" any item of interest? Is there a limit on the motion of the radar units? How many items need to be tracked?

-Predictor
http://will.dwinnell.com

There is no specific limit of how many items to chase, there might be some limits to where the sensors can move to, but that isn't very important right now.

What I need is an algo. that gives me the "prefered" placement every time I (the program) ask for it considering what my area of interest is and which objects is previously detected and inside this area.

As I said, this is part of my final year project at my college and therefore the algo should be something that's been *used before*. It should also preferably have been published somewhere.
My intentions is to compare my own algo. with some "industry standard" or whatever...

I don't know of any "standard" algorithms for this, although it seems like a natural fit for fuzzy logic. There are also spatial sampling algorithms which solve a vaguely (!) similar problem. Is there a "shape" to the radar coverage?

-Predictor
http://will.dwinnell.com

Ok, sorry, I didn't really mean standard, but something similar that has been *used before* (in some game for example) so I can use it for comparison in my work.
What spatial sampling algorithms were you refering to? Do you know the name or where I can find info about those?

I was thinking to use just simple sensors (radars) with a specified coverage-radius, (not very important).

Ok, sorry, I didn't really mean standard, but something similar that has been *used before* (in some game for example) so I can use it for comparison in my work.
What spatial sampling algorithms were you refering to? Do you know the name or where I can find info about those?

I was thinking to use just simple sensors (radars) with a specified coverage-radius, (not very important).

Please remember that I said this was vaguely related...

One important problem in spatial statistics is selection of spatial samples for kriging (a type of spatial modeling). As an example, think about the problem of finding the best places to drill for water (or oil, gas, etc.). Assuming that the distribution of water has some pattern, the problem is how to find the best locations to drill. Theoretically, one could drill a bunch of holes (in some problems, called "boreholes") all over the place, perhaps in a big grid, to find the best spot. Drawing samples (drilling holes) has an associated, non-trivial cost, however. In a quick search, I didn't find much material on this, but if this is of interest, try searching on things like:

spatial statistics
geostatistics
kriging
borehole(s)
well(s)

This whole line of thinking aside, you might establish as your benchmark simply some scattering (random, quasi-random or tesselated) without regard to the detection or movement of targets.

-Predictor
http://will.dwinnell.com