Granularity-Dependent Roughness Metric for Tactile Sensing Assessment

It is challenging to have a unified roughness metric representing the tactile sensation given the complexity of surface properties such as hardness, viscosity, and texture. In this article, a metric is proposed to evaluate tactile roughness across materials based on the granularity commonly used for sandpapers and implemented through a combination of sensing hardware, operating paradigm, reference object, and assessment method. A laboratory-customized FingerVision imitating the properties of human skin and mechanoreceptors is used as the core component of hardware. A wedge contact method is introduced as the operating paradigm to make full use of the array sensor and validated by a recognition task of 13 sandpapers. Moreover, the LSTM network is applied to a more challenging task and outperforms the conventional methods, indicating its superiority for temporally correlated samples. So that with the sandpapers' set being reference objects, an S2SR LSTM network is used as the assessment method and tested on sandpapers' set itself with the standardized RMSE of 0.0158. It is further tested on the fabrics with the correlation coefficient between the metrics and their linear mass densities being over -0.9740. Finally, it is applied to unidentified objects across materials, and the result is consistent with human tactile perception. The experimental results demonstrated that the proposed granularity-dependent roughness metric for a certain object is stable and repeatable, indicating that it is credible, valid, necessary, and valuable for tactile sensing assessment, paving the way to a unified tactile roughness representation for robot dexterous manipulation.