Omnidirectional Humanoid Walking using Model Predictive Control

Humanoid walking is considered one of the hardest problems in Robotics. Current state-of-the-art humanoid robots are able to achieve high speeds on flat ground. However, they still exhibit agility, dexterity, robustness, flexibility and energy efficiency far below than a typical human does. In this work, we present an omnidirectional walking pattern generator for a humanoid robot. We follow an approach based on the Zero Moment Point (ZMP) concept, which provides an useful criterion for biped stability. To avoid dealing directly with the complex dynamics of a high degrees of freedom humanoid robot, we used the 3D Linear Inverted Pendulum Model (3DLIPM) to approximate the robot dynamics. The resulting dynamics is linear, which allows us to use a linear Model Predictive Control (MPC) scheme. Finally, experiments results are shown to validate the method presented.