Balance control and locomotion planning for humanoid robots using nonlinear centroidal models

Abstract

Balance control approaches for humanoid robots have traditionally relied on low-dimensional models for locomotion planning and reactive balance control. Results for the low-dimensional model are mapped to the full robot, and used as inputs to a whole-body controller. In particular, the linear inverted pendulum (LIP) has long been the de facto standard low-dimensional model used in balance control. The LIP has its limitations, however. For example, it requires that the robot's center of mass move on a plane and that the robot's contact environment be planar. This thesis presents control and planning approaches using nonlinear low-dimensional models, aimed at mitigating some of the limitations of the LIP. Specically, the contributions are: 1) a closed-form controller and region of attraction analysis for a nonlinear variable-height inverted pendulum model, 2) a trajectory optimization approach for humanoid robot locomotion over moderately complex terrain based on mixed-integer nonlinear programming with a low-dimensional model, and 3) a quadratic-programming based controller that uses the the results from these low-dimensional models to control a simulation model of the Atlas humanoid robot.