Hybrid Position/Torque Ankle Controller for Minimizing ZMP error of Humanoid Robot

Abstract

The linear inverted pendulum model (LIPM) has been widely used to generate the walking patterns of humanoids. However, because LIPM is a simplified model of the humanoid, the position error between the real zero-moment point (ZMP) and the desired ZMP occurs. The task level of trajectory such as the center of mass (COM) trajectory is modified to reduce the ZMP error, but it reacts slowly. Therefore, it is more effective to use the torque of the humanoid ankle joint to respond to sudden disturbances. To minimize the error between the real ZMP and the desired ZMP, we propose the ankle hybrid position/torque controller. The proposed hybrid ankle controller consists of the joint position controller to follow the walking patterns and the joint torque controller to minimize the error of the ZMP. In particular, the momentum-based disturbance observer (MOB) is used to estimate the joint torque that is used for feedback in the torque controller. Therefore, the proposed ankle hybrid controller can be used without using the joint torque sensor. The proposed hybrid control reduced the ZMP error while following the walking patterns, even in the presence of an external force. The performance of the proposed hybrid ankle controller was verified in the MuJoCo simulator using the humanoid robot TOCABI.