Solution:

Place the region as $[0,1]^2 \cup \bigl([0,1] \times [1,2]\bigr) \cup \bigl([1,2] \times [0,1]\bigr),$ so it is the $2 \times 2$ square with the top-right unit square removed. Both coordinates of the midpoint are averages of numbers in $[0, 2],$ so the midpoint always lies in the $2 \times 2$ square; it fails to lie in the region exactly when it lands in the missing square, i.e. when $x_A + x_B \gt 2$ and $y_A + y_B \gt 2.$

If neither point is in the right square, then $x_A + x_B \le 2;$ if neither is in the top square, then $y_A + y_B \le 2.$ So failure requires one point in the top square and the other in the right square, which happens with probability $2 \cdot \frac{1}{3} \cdot \frac{1}{3} = \frac{2}{9}.$ In that case, one $x$-coordinate is uniform on $[0,1]$ and the other on $[1,2],$ so $x_A + x_B \gt 2$ with probability $\frac{1}{2},$ and independently $y_A + y_B \gt 2$ with probability $\frac{1}{2}.$

The failure probability is $\frac{2}{9} \cdot \frac{1}{4} = \frac{1}{18},$ so the desired probability is $\frac{17}{18}$ and $m + n = 17 + 18 = 35.$