Mapping the Pseudo-Force Illusion: Learning of Multidimensional Haptic Latent Functions

Pseudo-forces induced when playing asymmetric vibrations hold a tremendous promise for design ungrounded haptic interfaces. With simple vibrotactile actuators, it becomes possible to provide users with a salient sense of directional force, and guide them through rich and continuous haptic environments. However, the mapping from vibration patterns to perceived pseudo-force magnitude is nonlinear, multidimensional, and user-specific, making exhaustive search impractical. We propose a sample-efficient, human-in-the-loop framework to estimate this latent function using a Pairwise Gaussian Process with a hybrid acquisition strategy balancing exploitation and exploration. A user study shows that well-calibrated utility maps can be recovered within 60 trials. By efficiently recovering the global perceptual function, our method enables continuous, personalized control of pseudo-force magnitude for ungrounded interfaces.