Tactile Sensor System Processing Based on K-means Clustering

Development of a touch-sensitive (sensate) skin for robotic manipulators would provide tactile feedback for fine-grained dexterous control of robots interacting with objects in their environments, a capability that has largely been missing with robotic systems developed to date. A sensate skin for robots would require integration of hundreds or thousands of minute force or pressure sensors, each producing a localized response. Interpretation and extraction of useful information from the sensate skin presents a key technical challenge. In this paper we present a technique for analyzing data from tactile sensor arrays based on K-means clustering. Using a simplified contact model, the procedure estimates both magnitude and location for impacts on the sensate skin surface. Furthermore, it robustly accommodates a variety of sensor array densities by interpolating across areas of sensor response, providing accurate results even between sensing elements.