Separation of tactile information from multiple sources based on spatial ICA and time series clustering

Tactile sensing is becoming an essential function for robots in daily life for safe and natural human-robot interaction. Tactile sensors for wide detection areas are required so that robots can interact naturally with their surroundings by touch, but they complicate the tactile situations of robots and cause difficulties of tactile information processing. The conventional methods are not able to deal with these difficulties because they only consider tactile interactions which are focused on single target in limited environments. For a better understanding of tactile interactions with one or more surroundings, we propose a framework which explains relations between observed tactile information and the sources so that robots can handle the difficulties of complicated tactile situations. and to interact with the tactile signal sources properly. Based on our proposed framework, the understanding of tactile situations between sources and robots is made possible by the separation of tactile signals into several groups of signals with different statistical properties. In this paper, we focus on a separation method based on spatial ICA and time series clustering so that the robot understands tactile information and properly interacts with the tactile signal sources. results from experiments that analyzed real tactile data confirm the effectiveness of the method by simplifying the mixing conditions of tactile information from multiple sources.