A Bayesian Trust Inference Model for Human-Multi-Robot Teams

Abstract

In this paper, we develop a Bayesian inference model for the degree of human trust in multiple mobile robots. A linear model for robot performance in navigation and perception is first devised. We then propose a computational trust model for the human multi-robot team based on a dynamic Bayesian network (DBN). In the trust DBN, the robot performance is the network input, the human trust feedback to each individual robot and the human intervention are the outputs (observations). The categorical Boltzmann machine is used to capture the multinomial distributions that model the conditional dependencies of the DBN. We introduce the expectation maximization (EM) algorithm for the model learning and personalization. A factorial form of the EM algorithm is adopted for the multi-robot system where each robot has its corresponding latent trust state in the human mind. Bayesian inference is conducted to find the trust states, i.e., the trust belief. Based on the inferred trust states, we further derive the formulation to predict human interventions for model validation. A simulated human-UAV collaborative search mission is conducted with humans-in-the-loop. The experiment results show the Bayesian trust inference model can infer the degrees of human trust in multiple mobile robots and also predict human interactions with relatively high accuracy (72.2 $$\%$$ ). These findings confirm the effectiveness of DBNs in modeling human trust towards multi-robot systems.