It's important for the robot to know where it is within the playing board. Once you know where you are and which way you're heading, you have no problems of crashing into any fixed things in the play area, like hitting walls etc. Also, with a simple "follow a route" algorithm, which stays on track by using a good localization method, it's pretty simple to pick up a lot of points while totally ignoring the other robot or any or the more advanced game techniques.
Having established that we need effective localization, we are now designing and building it - below I describe how our current ideas are developing.
Our plan is to use visible or infrared lights on the fixed beacons. These lights will flash or change colour to make them easier for us to track and seperate them from the other lights around the arena, which could be considerably brighter than ours. We found that with most webcams the dynamic range of the colours in the image is pretty narrow - for example a very bright light like the sun appears exactly the same colour as a white wall (ie. both #FFFFFF, the brightest white the camera can represent). This was a problem for us because lots of other things in the arena would be detected since they were the same colour as light sources. Also, the camera blurred fast moving lights into lines, which again makes detection hard.
To compensate for this we modified the linux driver for the camera to reduce the exposure control dramatically (to about 1 16,000th of the original exposure time per frame) so the image only showed light sources and the rest of the image as black. this had a nice side effect of getting rid of all the blur as well.
