Investigating the use of low-cost tactile sensor in emulating mechanoreceptor patterns and in hardness-based classification

Tactile sensors and material classification are important aspects of improving robotics grasping. Previously many different tactile sensors and classification have been introduced. Tactile sensors used were either customized in different arrays or have complex architecture. Artificial mechanoreceptor type sensors have also been used to perform different types of classification, but hardness-based classification has never been investigated. In some cases, general tactile sensors were used to perform hardness-based classification, but it didn't show up good accuracy score while using different machine learning algorithms. This approach uses off shelf tactile sensors and emulates mechanoreceptor patterns using neural-spike encoder techniques to represent that off shelf tactile sensors have capability to emulate spike train which may indicate some patterns same as human receptors. And further using raw sensor data and digital spike patterns data as additional feature to performs analysis. Outcome of hardness-based classification with 20% test size data indicates that spike patterns data can increase predictability accuracy score in different combination of sensors data where with three sensors it remains highest of 93.9% improved from 91.8%. And spike patterns for some mechanoreceptors also showcase that off shelf can be used to generate mechanoreceptor patterns.