I made a magnetic levitation device once - it would suspend iron nail under electromagnet. It seems deceptively simple - when nail goes up, turn off magnet, when nail goes down, turn it on - but if you do so you get rapidly increasing oscillations - you have to have a circuit that blends in first and second derivatives of the position into the control signal. A great deal of complexity for a very simple unstable system.

Tangent: A similar problem was described in Sebastian Thrun's Udacity CS373 couse with respect to steering a self-driving car. It seems the similar principles should apply, except that the distance from the magnet will change the effect of gravity on the nail, which makes the problem more complicated.