Kernel Principal Components Analysis of Lateral Force Data for Quantitative Evaluation of 3D Features Rendering Fidelity

Haptic rendering technology is increasingly oriented toward representing realistic interaction with the physical world. A parallel challenge is how to achieve objective and quantitative evaluation of haptic rendering fidelity. This article addresses this issue, and involves kernel principal components analysis (KPCA) to determine a quantitative indicator for assessing rendering fidelity of 3D features. The KPCA approach explores eigen-subspace properties of two lateral force sets for respective real and virtual 3D objects, and obtains the quantitative evaluation indicator (spatial similarity) by computing the inner product of two kernel principal features. Finally, the proposed KPCA approach is applied to assess electrostatic tactile rendering ability of 3D sinusoidal humps on touchscreens, and it is also compared with the linear principal components analysis (PCA) method and perceptual ratings. Significant analysis results show that the spatial similarity with the KPCA approach is significantly higher than that with the linear PCA method, and that there is no significant difference in the averaged similarity with respect to the KPCA approach and perceptual ratings.