In this paper, we perform a data driven analysis of human perception of 2 D haptic data, and study possible structures of the perceptual deadzone. We describe an experimental set up where a user is subjected to a 2 D piecewise constant haptic force, and is asked to respond with a click whenever he/she feels any change in the stimuli. The jumps of the force vector are distributed in the first quadrant of a circle of radius 2 N. The timespacing between the jumps is a uniform random variable over a range of [2,3] seconds. A jump is labelled as perceived or nonperceived based on the clicks. After recording the haptic response of a user, we identify signal features and apply several classifiers to predict the user response. Our thesis is that a classifier which predicts better, gives a good structure of the deadzone. We start with the Weber and level crossings classifiers to predict the perceivability of a jump, and find that level crossings performs significantly better than the Weber classifier over a range of [0,2] N. The level crossings classifier gives the best fit circular deadzone. Then, we apply a general conic section based classifier which points towards an elliptical deadzone. The elliptical deadzone is studied in detail to study the directional behaviour. It is found that the accuracy of both the circular and elliptical deadzone are nearly identical for all the users. Additionally, eccentricity and orientation of the elliptical deadzone match well with that of the distribution of the input data. It signifies that a user does not have direction preference while perceiving the change in 2 D haptic force. Importance plot of the features used in a random forest classifier also confirms our observation.
data analysis,pattern classification,piecewise constant techniques,2D haptic data,2D kinesthetic perception,2D piecewise constant haptic force,Weber classifier,data driven analysis,deadzone analysis,human perception,level crossings classifiers,Deadzone,Haptics,Holdout cross-validation,Level crossings,Random forest,Simulated annealing,Weber classifier
Computer vision,Eccentricity (behavior),Computer science,Algorithm,Uniform distribution (continuous),Artificial intelligence,Classifier (linguistics),Jump,Conic section,Random forest,Haptic technology,Piecewise