Study of variation in human upper body parameters and motion for use in robotics based simulation

Abstract

This paper reviews the variations in human upper body motion of subjects completing activities of daily living. This study was completed to serve as a reference to evaluate the quality of simulated of human motion. In this paper we define the variation in motion as the variation in subjects' parameters (link lengths), joint angles, and hand positions, for a given task. All of these variations are related by forward kinematic equations. Motion data from eight healthy right hand dominant adults performing three activities of daily living (brushing hair, drinking from a cup, and opening a door) were collected using an eight camera Vicon motion analysis system. Subject parameters were calculated using relative positions of functional joint center locations between segments. Joint angles were calculated by Euler angle rotations between body segments. Hand position was defined as the origin of the hand frame relative to the pelvis frame. The variance of recorded human motion was analyzed based on the standard deviations of subject parameters, joint angles, and hand positions. Variances in joint angles were found to be similar in magnitude to root mean squared error of kinematics based motion simulation. To evaluate the relative variance, the forward kinematic solutions of the trials were found after removing subject parameter variance and reducing joint angle variance. The variance in the forward kinematic solution was then compared to the recorded hand position variance. Reductions in subject parameter and joint angle variance produced a proportionally much smaller reduction in the calculated hand position variance. Using the average instead of individual subject parameters had only a small impact on hand position variance. Modifying joint angles to reduce variance had a greater impact on the calculated hand position variance than using average subject parameters, but was still a relatively small change. Future work will focus on using these results to create formalized procedures for quantifying the human likeness of artificial human motions, to serve as a basis for performance comparison between different methods.