If you don't know what is wrong with the Roomba, then that would be your first step - there may be something severely wrong with it that would prevent you from using it.
If I was using a Roomba (in fact, I have a couple of "dead" ones, but I -do- know what is wrong with them) in such a fashion, and I didn't know where the problem was, or why it was "dead", I would gut it.
There's a lot of parts in the Roomba, but providing you want a robot platform and not a Roomba, you can gut a lot of it, and end up with a pretty nice platform with several Sharp IR sensors, the wheel motors with integrated encoders, the front wheel encoder/sensor (if it is a newer model, I think); get rid of side brush and cleaning head module (maybe pull the parts out for future use).
Basically, you want to reduce it to motors and sensor, then trace the wires from the connectors to the motors/sensors (so you don't hook anything up wrong), and try to power the motors with a battery to verify that they work (don't hook things up wrong to the wrong wires; you'll burn out the encoders - you want those).
I would stay away from piggybacking on the CPU board of the Roomba, and roll-my-own on everything else (you'll need a dual-h bridge driver for the motors, your own battery pack, and read/code for the sensor outputs, etc). But that is just me.
But before anything else, you need to either get the Roomba working enough again to know what is wrong/broken - or gut it to a "bare bones" platform (and test the parts along the way to know what works and what doesn't).
Lastly - I take it you have experience with this kind of thing, have the appropriate tools, and have done all the examples and such for the Arduino and understand that portion? If not, you might be biting off more than you can chew...
