Learning to Walk Dynamically: Optimal Control Parameterization

Most of the existing control schemes for walking bipeds are based on the so-called zero-moment point (ZMP) stability condition. Such schemes have three common characteristics: 1) bending at knees thus to avoid singular positions; 2) closed-loop controllers for trajectory tracking at the joints, 3) using higher power actuators, in particular at ankles, to compensate for the rotational motion (falling down) of the body above the ankle; This concept is in contrast with the concept of passive dynamic walking and cannot lead to energy efficient and naturally looking locomotion.