Identifying Human Hand Orientation around a Cylindrical Handlebar for physical Human-Robot Interaction

This paper is concerned with identifying the orientation of the human hand relative to a cylindrical handlebar. In physical Human-Robot Interaction, a handlebar is commonly used as the point of contact between the human operator and the robot. Identifying the orientation of the operator's hand provides the robot with additional information on how the operator interacts with the robot. A flexible sensor array composed of 160 pressure sensing cells was wrapped around a cylindrical handlebar. Grasping patterns of ten subjects was recorded. Support Vector Machine (SVM) and Bayesian Inference classifiers were implemented to identify the hand orientation of a subject relative to the handlebar. Principal Component Analysis (PCA) was used to reduce the number of features in the classification. Comparisons between the classifiers of SVM and Bayesian Inference, with/without PCA, were conducted for evaluating their accuracy. Two scenarios were used in the comparisons: in the first scenario, the training data and the test data were different but from the same subject; in the second scenario, the training data and the test data were from different subjects.