An Application of Machine Learning to Predict Stiffness Discrimination Thresholds Using Haptics

Abstract

The effectiveness of our interaction with the computer-generated environments is subject to our physical limitations in real life such as our ability of discriminating differences in stiffness or roughness. This ability, represented by Weber fractions, is usually quantified by means of psychophysical experimentation. The experimentation process is tedious and repetitive as it requires the same task to be completed by participants until the mastery at a certain stimulus level can be ensured before moving onto the next level. Moreover, these thresholds are dependent on the tested standard stimulus level and, therefore, need to be identified by separate experiments for every possible standard stimulus level. The purpose of the current study is to reduce the amount of experimentation and predict the thresholds for stiffness discrimination of individuals after being tested at a single stimulus level. The prediction models tested provide a moderate level of prediction power, but more features, potentially physical and demographical in nature, are needed to increase their effectiveness. The procedure described herein can be extended to any modality other than stiffness and, therefore, has the potential to predict overall palpation effectiveness of an individual after a feasible amount of data is obtained through experimentation.