Implementation of Convex Optimization Control on the MIT Humanoid
Yanran Ding, Elijah Stanger-Jones and Sangbae Kim
Abstract
This paper summarizes the practical aspect of the application of a convex optimization control on the MIT Humanoid. The hierarchical control framework consists of a model predictive control (MPC) with reduced-order model, followed by a reactive whole-body control (WBC) with full-order model. Our work emphasizes the critical components, including automatic differentiation, inverse kinematics and multi-threading, that enabled successful simulation-to-real transfer to the hardware. Preliminary experimental results are presented, including push recovery and stable walking on the treadmill with a maximum walking speed of up to 0.4 m/s.