Object judgment with an event-driven MEMS-LSI integrated tactile sensor array system

Abstract

We have developed tactile sensor systems for next-generation robots. To install a large number of tactile sensors, we have proposed micro electro mechanical systems-large-scale integration (MEMS-LSI) integrated tactile sensors. The integrated device has the following features: capacitive type three-axis force sensing, embedded diode-based temperature sensing, signal processing for sensing data digitalization, and event-driven response for efficient serial bus communication. This paper demonstrates a sensor array system as up-to 40 integrated tactile sensors which are connected on one bus line. After acquiring the sensing data from the sensor array system, we applied a machine learning technique for target object judgment. The objective of the judgment is to classify the targets into normal object and abnormal object. With the sensor array system, data preprocessing and tuned Recurrent Neural network (RNN)/Long Short-Term Memory (LSTM) neural network models, we achieved high-accuracy, high-precision, and high-recall scores for the experiment of the judgment.