A Soft Neuromorphic Approach for Contact Spatial Shape Sensing Based on Vision-Based Tactile Sensor

Robots with tactile sensors can distinguish the tactile property of the object, such as the spatial shape, in many robotic applications. The neuromorphic approach offers a new solution for information processing to encode tactile signals. Vision-based tactile sensing has gradually attracted attention in recent years. Although some work has been done on proving the capacity of tactile sensors, the soft neuromorphic method inspired by neuroscience for spatial shape sensing is remarkably rare. This paper presented a soft neuromorphic method for contact spatial shape sensing using a vision-based tactile sensor. The outputs from the sensor were fed into the Izhikevich neuron model to emit the spike trains for emulating the firing behavior of mechanoreceptors. 9 spatial shapes were evaluated with an active touch protocol. The neuromorphic spike trains were decoded for discriminating spatial shapes based on k-nearest neighbors (KNN). Three spike features were used: average firing rate (FR), the coefficient of variation of the interspike interval (ISI CV), and the first spike firing time (FST). The results demonstrated the ability to classify different shapes with an accuracy as high as 93.519%. Furthermore, we found that FST significantly improved spatial shape classification decoding performance. This work was a preliminary study to apply the neuromorphic way to convey the tactile information obtained from the vision-based tactile sensor. It paved the way for using the neuromorphic vision-based tactile sensor in neurorobotic applications.