AutoBalancer An Online Dynamic Balance Compensation Scheme for Humanoid Robots

Overview

Satoshi Kagami, University of Tokyo, Tokyo, Japan

Fumio Kanehiro, University of Tokyo, Tokyo, Japan

Yukiharu Tamiya, Namco Ltd, Tokyo, Japan

Masayuki Inaba, University of Tokyo, Tokyo, Japan

Hirochika Inoue, University of Tokyo, Tokyo, Japan

Algorithms for maintaining dynamic stability are central to legged robot control. Recent advances in computing hardware have enabled increasingly sophisticated physically based simulation techniques to be utilized for the offline generation of dynamically-stable motions for complex robots, such as humanoid robots. However, in order to design humanoid robots that are reactive and robust, a low-level online balancing scheme is required.

This paper presents an online algorithm for automatically generating dynamically stable compensation motions for humanoid robots. Given an input motion trajectory, the "AutoBalancer" software reactively generates a modified dynamically-stable motion for a standing humanoid robot. The system consists of two parts: A planner for state transitions derived from contacts between the robot and the ground, and a dynamic balance compensator which formulates and solves the balance problem as a constrained, second order nonlinear programming optimization problem. The balance compensator can be made to compensate for deviations in the centroid position and tri-axial moments of any standing motion for a humanoid robot, using all joints of the body in real-time. The complexity of the Auto-Balancer algorithm is O((p + c) ³), where p is number of DOFs and c is the number of constraint equations.

We describe results obtained by an experimental implementation of the AutoBalancer algorithm that has been applied to 16-DOF and 30-DOF...