Estimation of unknown node positions of a localization network with a multi-robot system

In this thesis, a novel method for estimating the node positions of a localization network is presented. A multi-robot system is used to map the positions of the network nodes, while the robots track their own position simultaneously. It is an application of simultaneous localization and mapping (SLAM). The localization is based on bearing angle measurements between a robot and a network node. Hence, the method used for the localization can be called bearing-only SLAM. The localization method is based on a probabilistic approach. All the measurement data are collected to a centralized Kalman Filter. As a result of the non-linear measurement equation, the Extended Kalman Filter (EKF) algorithm is used. The centralized structure maintains the covariances between all the entities and thus takes full advantage of the cooperation in a multi-robot system. The algorithm is shown to work with a sparse distribution of landmarks. A robot makes a bearing angle measurement to only one landmark at a time. Therefore, the computational complexity of the Kalman filter stays low. The Radio Frequency Identification (RFID) technology is used in the case study presented in this thesis. It is shown that passive RFID tags can serve as landmarks with a unique ID. The inexpensive, maintenance-free RFID tags can easily be distributed over the intended working area of the robots to form a localization network. The bearing angle measurements to the RFID tags do not need to be highly accurate as the proposed algorithm can handle uncertain measurements. Simulations and laboratory experiments are used in order to prove the performance of the proposed method.