Measuring motion expressiveness in wheeled mobile robots

This paper addresses the measurement of motion expressiveness in wheeled mobile robots. A neural network based supervised learning strategy is proposed as a method to fuse information obtained from the measurement of selected features. The choice of these features is made to reflect the visual quality of the trajectory and hence carries semantic ambiguities that are filtered out through the ability to generalize knowledge by the neural network. The paper presents results with two features that might be significant in what concerns motion expressiveness, namely, how confident/hesitant is the motion and whether or not contains local loops that might indicate, for example, a call for attention by the robot towards a group of humans.