maybe a screen shot or two would help better explain the problem, but in the meantime have you tried playing around with the tolerance on the rigid solver?