Statistical analysis of haptic data

Abstract

The science op applying touch (tactile) sensation and control to interaction with computer applications is called haptic. In this study interest was to measure the effect of haptic sensation on the evaluation of virtual prototypes. In this context, forces are generated and felt by the user and this is done hrough haptic algorithms. The aim of this report was to investigate which statistics should be used to compare algorithms. The data is collected from an experiment conducted in the laboratories of University of Hasselt more precisely at the EDM (Expertise Centrum for Digital Media). Three virtual objects(an orb, fish and a cube) were used. First a reference algorithm is used where the position and velocity are recorded, secondly a set of 6 algorithms are used producing outputs on which comparisons can be made. Each algorithm is used on each object 50 000 times resulting to 900 000 observations in total. Performance of each algorithm was characterized using execution time and the surface contact points (SCP). Analysis of variance (ANOVA) was implemented to the data set in order to investigate differences between all 6 algorithms. Multivariate normal regression was also implemented to investigate the performance based on both the SCP and execution time together, thereby taking into account the correlation between the multiple responses. Due to departures from normality, generalized estimating equations with a gamma distribution for the errors were further applied. All models indicated that the algorithms performed differently, and the performance depends on the object on which the algorithm is being applied on.