Modeling Trust Dynamics in Human-robot Teaming: A Bayesian Inference Approach

Abstract

In this work, we proposed a personalized trust predictor for modeling trust dynamics in human-robot teaming. The proposed method models trust by a Beta distribution to capture the three properties of trust dynamics, which takes the performance-induced positive attitude and negative attitude as parameters. The model learns the prior distribution of the parameters from a training dataset, and estimates the posterior distribution based on a short training session and occasionally reported trust feedback. The experiments showed that the proposed method accurately predicted people's trust dynamics, achieving a root mean square (RMS) of 0.0724 out of 1.