Two-Dimensional Direction Recognition Using Uniaxial Tactile Arrays

To allow intuitive communication in human-robot cooperation through tactile information, this paper presents a method to recognize human intended direction in 2-D using a handlebar equipped with uniaxial tactile arrays. The method first extracts various features from the tactile images aiming to reduce computation complexity and increase recognition robustness. A support vector machines classifier was implemented for classifying the intended direction of humans using the extracted features. The algorithm efficiency of using different combinations of features has been investigated and compared through human user studies. In total, five human users in the project team were involved in this research. Experimental results show that the proposed method can achieve 91.7% recognition accuracy if both the training data and validation data contain tactile images from all the users. The method could still achieve 77.5% recognition accuracy when the training and validation data share no common user.