Cross-Talk Compensation in Low-Cost Resistive Pressure Matrix Sensors

For socially assistive robots in close contact to people, a tactile sensor can be useful for gathering feedback and inputs in the form of touch gestures. In this paper, we concentrate on low-cost textile pressure matrix sensors since they are easy to manufacture and due to their flexibility can be adopted to the curved shape of a robot's outer cover. Due to the matrix principle for reading out, the setup suffers from artifacts when it comes to activation of multiple sensor elements. We present a machine learning approach for preprocessing the raw measurements from the pressure sensitive array in order to get reliable pressure patterns which can be used for gesture classification later on. By means of that, an expensive hardware solution for capturing the pressure values can be avoided.