Rather than trying to track multiple things externally, why not have each thing determine where it is in the room, based on it's surroundings and past experience?
Think about it: If you are in a room, you can tell where you are in that room fairly accurately; how far away from the walls and objects you are, what direction you are facing or about to head in, etc. You could then easily communicate this to others in the room, or to a "central" system (via a radio, perhaps).
So why not try to emulate this with each robot or device in the room. It could take a measurement with its sensors (it's going to need more than one type of sensor, by the way, if you are going the inexpensive route - ideally, you would use a LIDAR system, but we are trying to keep costs down - so maybe a combo sonar and Sharp IR sensor would work well, perhaps mounted on a rotating/scanning servo?), then take a guess at where it is at on an internal map (hint: it's first guess will be waaaaay off!), then it could move, then take another measurement, then update its guess, and so on. The more it moves around and measures, the greater it's probability of knowing where it is at in the environment. It will -never- know -exactly- where it is, but it -will- know, within a certain level of probability, close to where it is! Believe me - this really works!
Of course - it isn't easy - in fact, the basic algorithm I described is very hard to implement (plus you need to deal with massive uncertainties about the senor(s) themselves, etc). It is something that still hasn't been completely solved; it is a very fertile ground for research and investigation, with numerous real-world applications. It even has a name:
Simultaneous localization and mapping - Wikipedia (aka SLAM)
That wikipedia article is a good first look at what SLAM is all about. Be sure to also check out the "SLAM for Dummies" link at the bottom of the article as well.
Finally, if you are really serious, I encourage you to participate in this online course as well:
http://www.udacity.com/overview/Course/cs373/CourseRev/apr2012
Note that the above course requires you to be somewhat familiar with linear algebra, statistics/probability, as well as Python (as the course uses Python for the programming portions). It is a free, work at your own pace course. Well worth it if you really want to learn the basics of SLAM, path planning, and a whole host of other Artificial Intelligence concepts...