Development of artificial haptic model for human tactile sense using machine learning

Recent studies have proposed a model using the supervised neural-network (NN) as the haptic model. However, this model has the problem in expressing human tactile sense. Therefore, we constructed a haptic stage model based on the model of human-tactile sense and cognitive process. Our NN-based model consists of supervised and unsupervised stages. Using surface-scanning results from our trial micro-tactile sensor, we confirmed the material-identification problem with our model. A conventional full conventional NN(CNN) model achieved a 98.0% correct identification rate for 28 materials; our model achieved a rate of 67.2%. Our model employed typically is deepening because we used the convolutional process in all layers, therefore, we consider that our learning result lost local features on each material, and then identification rate was declined as result.