An adaptive controller based upon the online minimization of a performance criterion is described. The adaptive controller is used to improve the performance of a one-legged mobile robot, removing problems experienced with previous controllers. Specifically, this controller eliminates the problem of a bias in the forward velocity experienced with the nonadaptive controller and, at the same time, provides the flexibility required to allow a dynamically stabilized legged machine to perform satisfactorily under the widely varying conditions that exist in the real world. The performance of several minimization algorithms is analyzed, and the adaptive step size random search algorithm is selected. Finally, a series of experiments illustrating the ability of the adaptive controller to handle a changing environment is presented.>