A model of the user's proximity for bayesian inference

Abstract

Embodied nonverbal cues are fundamental for regulating human-human social iteractions. The physical embodiment of robots makes it likely that they will have to exhibit appropriate nonverbal interactive behaviors. In this paper we propose a model of the user's proximity based on a superposition of quasi-Gaussian probability distributions which allows to express findings from HRI trials regarding distances and direction of approach in a human-robot interaction scenario. The way the model is formulated is suitable for well-established Bayesian filtering techniques, and thus the inference of the preferred distance and direction of approach in a human robot interaction scenario can be regarded as a state estimation problem. Results derived from simulations show the effectiveness of the inference process.