Kneed passive dynamic walker controller design employing a climbing learning algorithm

How to design a controller for the kneed passive dynamic biped walker is a challenge due to the complicated dynamics. In this paper, a controller is designed to build the kneed passive dynamic biped robot to start walking from a random state with zero velocity at anytime, and after starting, the biped robot should enter into the limit cycle smoothly. In order to achieve this objective, a climbing learning algorithm cooperated with the PD control is proposed. The climbing learning algorithm is to learn the PD parameters in the PD control. The target function in the climbing learning can be obtained online in the learning process, and the function is the evaluation of the PD control with the given PD parameters. The experiments and studies show that the kneed passive dynamic walker can converge to the limit cycle quickly with the learned controller.