Embedded Implementation of Signal Pre-processing for Tactile Sensing System

Smart tactile sensing system has been a subject of research in many application domains such as prosthetics and robotics. Embedding signal pre-processing methods (i.e., filters) along with processing algorithms (i.e., machine learning) into miniaturized electronic units enhance the extraction of high-bandwidth information (e.g., slippage detection). However, it is challenging due to the high computational costs and the real time requirements. This paper proposes a lightweight implementation of pre-processing method for multichannel tactile sensing system. We targeted two filtering methods, Finite Impulse Response (FIR) and Exponential Moving Average Filter (EMAF). The paper presents the analysis of the implementation performance on hardware i.e., number of clock cycles, execution time and touch detection accuracy. Experimental results show that EMAF is more effective than FIR when it comes to the hardware complexity. This means that the computational cost for implementing such pre-processing filter is negligible and thus acceptable for time, and hardware constraint tactile sensing system.