Optimization of Pelvic Rotation Walking Pattern Considering Future States Using Model Predictive Control to Increase the Step Length

Abstract

Pelvic rotation, which is observed in human gait, is used to increase a robot’s step length in the humanoid robot walking. Existing methods empirically or experimentally generate the pelvic-rotation angle offline using predetermined pelvis and foot trajectories. Therefore, these methods are difficult to be used with the method to generate the center of mass (CoM) trajectory in real time using techniques such as preview control or model predictive model control. In this study, we propose a method that generates a pelvic-rotation trajectory that can be used with a real-time CoM generation method while reflecting the future state of the robot. The step length increased due to pelvic rotation was kinematically analyzed and compared with the step length increased by the proposed method. Upper body motion to compensate for yaw angular momentum generated by lower body movement during walking using pelvic rotation was optimized using centroial dynamics. The waist yaw joint for pelvic rotation, leg joint for the walking, and upper body joint for arm swing were optimized using separate optimal controllers. Energy efficiency, increase in step length, and decrease in the possibility of singularity occurrence of pelvic rotation walking were analyzed and compared in simulations. The proposed method was experimentally verified using humanoid robot Dyros-Jet.

Keywords

• Humanoid robot
• locomotion
• trajectory optimization
• model predictive control
• quadratic programming