Stiffness in Virtual Contact Events: A Non-Parametric Bayesian Approach

In this study we investigated the use of simple vibrotactile signals to simulate contact with a virtual object. In particular we explored the relation between properties of the signal and the perceived hardness of the object. The space of stimuli is large, and we have no plausible a priori model for the relationship of parameters to percept. Thus we made use of non-parametric Bayesian methods, in particular utilizing Gaussian process priors. We show that this method both gives insight into the phenomenon of interest and well-predicts a second, separate data set collected via the method of constant stimuli. Thus we argue that it could be a fruitful approach for attacking a variety of perceptual problems.