Designs and Evaluations of Environmental Change Detection Methods Based on Statistical Decision Theory

Abstract

This paper deals with parameter estimation problems on dynamic environments. To deal with environmental changes, autonomous mobile robots need to estimate parameters that are related to the changes. To estimate such parameters, a possible approach is to estimate the breakpoint first, and then use it to estimate the parameters. However, such a way of estimation may lack of the optimality that is argued in statistical decision theory. That is, it may not minimize the expected loss that consists of the loss function that is consistent with an evaluation criterion. Therefore, this paper numerically evaluated how much the performance of the estimators that are formalized from each standpoint can differ.