I. Introduction
U nderactuated balancing systems such as bicycles or motorcycles provide a unique platform to train and treat human with postural balance disabilities [1]. In [2], an actively controlled bicycle-based robot, called bikebot, is developed to perturb or assist rider's sensorimotor mechanisms. Balance control of the bikebot is however presented only for stationary case in [2]. To further understand and tune the human balance motor skills in these unstable physical human-robot interactions [3], [4], it is desirable to design an autonomous trajectory tracking and balance capability for the bikebot.