Human fingertip tactile perception relies on the activation of densely distributed tactile receptors to identify contact patterns in the brain. Despite significant efforts to integrate tactile sensors with machine learning algorithms for recognizing physical patterns on object surfaces, developing a tactile sensing system that emulates human fingertip capabilities for identifying contact patterns with a high spatiotemporal resolution remains a formidable challenge. In this study, we present the development of an artificial tactile finger for accurate contact pattern identification, achieved through the integration of a high spatiotemporal piezoresistive sensor array (PRSA) and a convolutional neural network (CNN) model. Spatiotemporal characterization tests reveal that the artificial finger exhibits a fast temporal resolution of approximately 7 ms and achieves a two-point threshold of 1.5 mm, surpassing that of the human fingertip. To compare the performance of the artificial finger with the human finger in recognizing different patterns, we acquired pressure images by pressing the artificial finger, coated with a flexible PRSA film, onto both simple embossed and complex curved patterns while also recording human recognition results of perceiving these patterns. Experimental findings demonstrate that the artificial finger achieves higher classification accuracy in recognizing both simple and complex patterns (99.0 and 96.1%, respectively) compared to the human fingertip (69.1 and 22.7%). This artificial finger serves as a promising platform with great potential for various robotic tactile sensing applications including prosthetics, skin electronics, and robotic surgery.
Keywords: convolution neural network; flexible tactile sensors; high spatiotemporal resolution; machine learning; pattern identification; piezoresistive sensor array.