Abstract

In smart environments, a multi-robot system is difficult to achieve a high confidence level of information for reliable decision making. The pieces of sensed information obtained from a multi-robot system have different degrees of uncertainty. Generated contexts about the situation of the environment can be conflicting even if they are acquired by simultaneous operations. In particular, unpredictable temporal changes in sensory information reduce the confidence level of information then lead to wrong decision making. In order to solve this problem, we propose a reasoning method based on Dynamic Evidential Fusion Network (DEFN). First, we reduce conflicting information in multi-sensor networks using Evidential Fusion Network (EFN). Second, we improve the confidence level of information using Temporal Belief Filtering (TBF) and Normalized Weighting technique. We distinguish a sensor reading error from sensed information. Finally, we compare our approach with a fusion process based on Dynamic Bayesian Networks (DBNs) using paired observations so as to show the improvement of our proposed method.