Analysis of affective human motion during functional task performance: an inverse optimal control approach

For robots that collaborate alongside and work with humans, there is great interest in improving robot communication abilities to achieve engaging and successful interactions. Successful task collaborations between humans often involve functional motions in which implicit communication signals, such as affect, are embedded. Thus in order to improve a robot's communication capabilities, it is necessary to identify the different motor control strategies that humans employ when generating such implicit signals. This paper details the adaptation of an Inverse Optimal Control (IOC) methodology for this purpose. We hypothesize that IOC allows for the identification of the motion strategies involved in the implicit communication of affective content during the performance of functional movement. To test our hypothesis, a motion capture dataset consisting of upper-body functional movements was collected and annotated by multiple observers through a perceptual user study. Among the different control strategies considered during our analysis, we found that center of mass movement, quantity of motion, Laban space effort and effort were the most relevant when distinguishing motions that convey different affective states.