We propose a method of shape classification in rotating manipulation by a multi-fingered robot hand. A multi-fingered robot hand understands position and posture of a object and can realize suitable manipulations by classifying a shape of contact surface in rotating manipulation. Our method uses pressure distributed sensors equipped in fingers and consists of the following three processes. First, periodic pressure distributions are measured by a pressure distributed sensor in rotating manipulation. A kurtosis is calculated from each pressure distribution and can quantify a shape of contact surface in the moment. Secondly, a kurtosis pattern is cut out from a periodic kurtosis. Finally, a degree of similarity calculates between a kurtosis pattern and a reference pattern by DP matching method and classifies the shape of contact surface using thresholds. We confirm that our method classifies three objects with a high degree of accuracy and can keep the classification rate even in changing rotation velocity through laboratory experiments.
