Do you have any restrictions on a? Is it integer or any real number?
I suggest you start with the definition of the expectation value... you sum or integrate over values times the probability distribution:
[tex]E(g(X,Y)) = \int_{x,y\in[0,1]} g(x,y)\cdot f(x,y) dxdy[/tex]
where f(x,y) is the joint probability distribution. Since X and Y are both uniform the joint distribution will be f(x,y) = constant. You can determine this constant by normalization:
[tex]\int_{x,y\in [0,1]} f(x,y) dx dy = 1[/tex]
As far as doing the integral involving the absolute value you should break the square over which you are integrating into the region where (x-y)>0 and (x-y)<0.