In a network robot system, a robot and a sensor network are integrated smoothly to develop their advantages and benefit from each other. Robot localization, sensor network calibration and environment mapping are three coupled issues to be solved once network robot system is introduced into a service environment. In this article, the problem of simultaneous localization, calibration and mapping is raised in order to improve their precision. The coupled relations among localization, calibration and mapping are denoted as a joint conditional distribution and then decomposed into three separate analytic terms according to Bayesian and Markov properties. The framework of Rao-Blackwellized particle filtering is used to solve the three analytic terms, in which extended particle filter is used for localization and unscented Kalman filter is used for both calibration and mapping. Simulations have been performed to demonstrate the validity and efficiency of the proposed solutions.