Iterative learning of human partner’s desired trajectory for proactive human–robot collaboration