Computer to robot is a bundle of worms all in itself.. basically the most common way to do this is to use "sub systems", which basically means you use MCU's with serial ports in order to control the smaller systems..
For example, in my balance bot I plan to use the basicAtom Pro-24m with an io board to control the balancing.. so it will take in the gyro's and acceleramators and do some quick calculations and then determine what motor commands to send to the h-bridge to control the motors. It will do this and just repeat it forever, every calculation loop or so it will read the serial buffer to check for commands. Such as go forward, back, side, side, etc etc. Or requests for data and stuff...
While lots of robots use only a single MCU to do everything.. using a mini-itx or laptop as the brain of the bot.. you can link together a few mcu's and diffirent things VIA serial ports and create a far more robust robot.. and also use the mini-itx to do more crazy things.. such as wifi, blueototh, image proccessing, gps, etc etc.. a lot of things are avialable for computers now, that are NOT around for robots yet..
I feel mini-itx/laptop options allow really great starts to robots.. but it is more expensive. more complicated.. and can be seriously difficult. As an example:
One might think that you can use the mini-itx to control quick responses such as bump sensors and such.. but the system is not meant for interrupts. You have let the intterrupts be controlled by MCU's typically.. while mini-itx is great for mapping and all