High frequency walking pattern generation based on preview control of ZMP

Abstract

This paper addresses a dynamically stable humanoid walking pattern generation system that can update the pattern at a short cycle such as 40 [ms]. Frequent update of the pattern enables a robot to reflect change of moving direction and upper body posture with short delay after the change of the command input. It also enables a robot to change the walking pattern according to the change of measurable applied force. Moreover short cycle generation is useful in constructing a system that reflects motion status modified by sensor feedback to the motion generation layer in order to maintain consistency of dynamics. Preview control theory is adopted to decide horizontal torso trajectory that realizes desired ZMP trajectory. The proposed system is implemented on a full-size humanoid HRP-2. Walking pattern is updated at every 40 [ms] with reflecting the current modified torso position in the system. Walking control is successfully carried out by the implemented system. Then an experiment of generating and executing walking pattern while variable force is applied to the hand is shown. The system generates walking pattern reflecting the force sensor information at the wrist while walking