EXAMPLE GUIDED INVERSE KINEMATICS

This paper proposes example guided inverse kinematics (EGIK) which extends and enhances existing inverse kine-matics technique. In conventional inverse kinematics, redundancy in the model produces an infinite number of solutions. The motion could be jerky depending on the choice of solutions at each frame. EGIK exploits the redundancy for imitating an example motion (a pre-measured motion data) so that a unique solution is chosen. To minimize the gap between the goal and current end-effector position and imitate the original motion at the same time, nonlinear optimization technique is employed. So, the resulting motion resembles the original one in an optimal sense. Experiments prove that the method is a robust and effective technique to animate high DOF articulated models from an example motion. 1 INTRODUCTION Animating human motion has been a great challenge. The task may appear easy at the first look since we can completely command an articulated figure by supplying joint angles. However, the difficulty stems from the fact that there are too many things to control. Human body has 206 bones and hundreds of muscles. A reasonable model of it can easily have 40 degrees of freedom. Computing such number of joint angles so that the resulting motion resembles that of a real human is not a trivial task. Among diverse approaches to solve this problem, inverse kinematics and motion capture are just two. The algorithm proposed in this paper is about the half way between these two approaches. Inverse kinematics was originated from robotics field [1]. It computes joint angles that position the end-effector at a desired location. In robotics, major interest has been on six DOF robots. Since end-effector has six DOFs in general (three for position, and the other three for orientation), inverse kinematics on a six DOF robot gives a unique or at most four different solutions. However, if a model has 40 DOFs, there exist an infinite number of solutions (actually, the dimension of the solution space is 34), and only one of them is selected for the frame. Because the selection is purely up to the numerical process employed, even though the end-effector follows anticipated trajectory, joint angles can make abrupt changes. Therefore neighboring frames won't have coherence, and simple replay of those frames may result in a jerky animation. Usually the numerical process picks a configuration that is reasonably close to the previous configuration. Therefore, in interactive demonstration, many times the …