If a communication robot could have the same capability of haptic interaction as humans do, the robot would give us greater familiarity, thus shortening its communicative distance from people. To achieve such a robot, this paper proposes a method for categorizing haptic interactions between a human and a robot by using tactile sensors embedded in the soft skin covering the robot's entire body. In this method, each haptic interaction can be represented by a sequence of data points recorded from the tactile sensors. We first consider all experimental data points and map them into clusters. Next, we represent each individual haptic interaction as a sequence of those clusters. Finally, by calculating a distance function between each of these sequences of clusters, we can determine a hierarchical structure that naturally groups similar haptic interactions into categories. In the hierarchical structure, the haptic interactions are classified mainly based on the region of touching (e.g. head and arm) at the top side of the structure and the manner of touching (e.g. slap and stroke) at the bottom side of the structure. By using this hierarchical structure, the robot can categorize haptic interactions based simultaneously on the region of its body touched and on the manner of touching. The constructed hierarchical structure will also make the robot aware of the categorized regions and the categorized manners of touching that are typically observed in haptic communication between a human and a robot. Consequently, this capability will lead to more human-like communication. To confirm the validity of this method, we conducted human-robot communication experiments using 284 subjects. As a result, we clarified that the manner of touching depends on the region of touching.
展开▼
