Measuring motor actions and psychophysiology for task difficulty estimation in human-robot interaction

Abstract

In this paper, we describe a method for estimating task difficulty in human-robot interaction using a combination of motor actions and psychophysiology. A number of variables are calculated from kinematics, dynamics, heart rate, skin conductance, respiration and skin temperature. Discriminant analysis of the variables is used to determine whether the user finds the task too easy or too hard. The discriminant function is recursively updated with Kalman filtering in order to better adapt to the current user. The method was tested offline in a task with 20 subjects. In cross-validation, nonadaptive discriminant analysis yielded a classification accuracy of 80.2% while adaptive discriminant analysis yielded a classification accuracy of 84.3%.