There's a little bit more on this in Chapter 6.3.8 of the Topas book Robert, Andreas and I put together (
http://topas.dur.ac.uk/topaswiki/doku.php?id=book).
I don't have an example to hand, but the idea is Alan describes in the wiki post you cited.
Essentially you have some initial guess of the sites (he called these wC1, wC2). You can put these at 0 occupancy so they don't later influence the fit tot the data.
You then have the "real" sites that are described by the rigid body (C1, C2).
You write a set of distance restraints so that C1 is close to wC1, C2 is close to wC2 etc.
You then use only_penalties to minimise the differences.
You then have your starting positions and angles for the rigid body.