Edge AI Algorithm for FBG-Based E-Skin Touch Localization on Embedded Electronics

The field of collaborative robotics is constantly growing given the central role that safe human-robot interactions are gaining in the current era of Industry 4.0. To act and interact, robotic systems are increasingly being either natively equipped with or adapted to house sensing devices. In this scenario, tactile sensors are fundamental to detect impacts, accurately manipulate objects and enable learning from demonstration strategies. The development of such systems requires technological efforts in terms of both design, for their seamless integration in robots, and software programming, for achieving reliable and prompt reactions to relevant events. In this study, an edge AI algorithm deployed on an embedded device for the recognition of contacts on an FBG (Fiber Bragg Grating)-based e-skin is presented. A Convolutional Neural Network has been trained offline to localize touch events occurred on the e-skin. The performance of the algorithm assessed on the test set reached a mean absolute error on the three cartesian coordinates of less than 4 mm. The algorithm has been then ported to a microcontroller board for real-time testing of both simulated and concurrent contacts. The achievements of this work pave the way to new opportunities for collaborative robotics applications through intelligent FBG-based artificial tactile skins.