Design of human-like posture prediction for inverse kinematic posture control of a humanoid robot

Abstract

A method and system has been developed to solve the kinematic redundancy for a humanoid redundant manipulator based on forward kinematic equation and the optimization of human-like constraints. The Multiple Objective Optimization (MOO) is preformed using a Genetic Algorithms (GA) and implemented using the Genetic and Evolutionary Algorithm Matlab Toolbox. The designed system is illustrated on a simple redundant 3 degree of freedom (dof) manipulator and is set up for a more complicated redundant 7 dof manipulator. The 7 dof manipulator is modeled from the Stan Winston studio's Leonardo, an 61 dof expressive humanoid robot. It has been found that the inverse kinematic solution to a 3dof model arm converged within 1% error of the solution within .05 mins processor time using the discomfort human-like constraint in 2d space. Similarly, the inverse kinematic solution to a 7 dof model arm consisting of Leonardo's right arm geometry was found to converge within 1% error within .20 mins processor time using the discomfort human-like constraint in 3D space. The full kinematic model of Leonardo is developed and future efficiency optimizations are posed to move towards the real-time motion control of a redundant humanoid robot by way of human-like posture prediction.