Learning Global Direct Inverse Kinematics

Abstract

We introduce and demonstrate a bootstrap method for construction of an inverse function for the robot kinematic mapping using only sample configuration-- space/workspace data. Unsupervised learning (clustering) techniques are used on pre--image neighborhoods in order to learn to partition the configuration space into subsets over which the kinematic mapping is invertible. Supervised learning is then used separately on each of the partitions to approximate the inverse function. The ill--posed inverse kinematics function is thereby regularized, and a global inverse kinematics solution for the wristless Puma manipulator is developed. 1 INTRODUCTION The robot forward kinematics function is a continuous mapping f : C ` Q n !W ` X m which maps a set of n joint parameters from the configuration space, C, to the m-- dimensional task space, W. If m n, the robot has redundant degrees--of--freedom (dof's). In general, control objectives such as the positioning and orienting of the end-- ...