On experimental considerations about discriminations of applied forces to a pressure sensitive conductive rubber with a neural network

Abstract

Recently, studies to realize a ``smart hospital'' are carried out by many researchers. To this end, robots are expected to learn palpation to act as doctors or helpers. To do that we need training data constructed by the magnitude of applied pressures and output data from a pressure sensor during palpation, which are difficult to be obtained with standard pressure sensors. To overcome this issue, we use a sensor with pressure-sensitive conductive rubbers(PSCR) and an estimation system with a neural network(LSTM) which uses the output data from the sensor as input and we study what is appropriate input data to the estimation system. In this paper, we show experimental results such that an estimated accuracy is increased by about 0.18 points using the first and the second derivative of the outputs.