Modeling Reach Motions Using Functional Regression Analysis

This article describes a method for predicting human motion where some part of the body, such as the pelvis or foot does not move. The posture at any given time can be approximated using a linkage of articulated segments. The angles between the segments describe the posture. During the reach, these angles will vary describing a function that varies over time. Data may be collected on individuals reaching to a variety of targets. We describe a functional regression model for predicting the angles as they change with time as a function of the target being reached, the anthropometry and other characteristics of the individual. The model may be used to predict the motion of new individuals reaching to new targets not contained within the data. The model can also be used to describe the effects of factors such as age on reaching motions. Although the main purpose of this article is to describe the methodology, we demonstrate its use on a large database of reaching motions collected by HuMoSim at the University of Michigan. We discuss the accuracy of the predictions and the difficulties ensuring that the hand reaches the intended target.