Real-Time Pressure Estimation and Localisation with Optical Tomography-inspired Soft Skin Sensors

Abstract

Sensing and localising pressure resulting from physical interaction between a robot and its environment is a key requirement in the deployment of soft robots in real-life scenarios. In order to adapt the robot's behaviour in real-time, we argue that sensors must have a high sampling rate. In this paper, we present a novel tactile sensing strategy for soft sensors, based on an imaging technique known as optical tomography. Instead of transmitting light through the soft sensor in a sequential way (as commonly done in tomography systems), we demonstrate that concurrently illuminating the sensor with multiple light sources and reading out the sensor response has several advantages. Firstly, it drastically increases the sampling rate of the sensor when compared to standard tomography approaches, making it more suitable to sense sudden and short-lived contacts. Secondly, by concurrently switching on the light sources, we increase performance in terms of pressure localisation and pressure estimation achieved through Machine Learning techniques. We carry out experiments demonstrating that our approach allows for a robust pressure estimation and contact point localisation with an accuracy up to 91.1 % (vs 70.3%) at a higher sampling rate.