Accurate localization of mobile robot under the weakly callibrated motion model

This paper proposes an accurate localization method, which consists of a non-probabilistic motion model and Generalized Iterative Closet Point (GICP). The most encountered problem of using motion models is to determine empirical parameters, which represent the systemic errors and the non-systemic errors. The perfect representation of those errors is an extremely difficult task, and it is practically impossible. Therefore, in order to compensate those errors, generally a probabilistic approach is used in the grid map. However, this generic approach shows a drifting problem due to systemic/non-systemic errors of the motion model and discrepancy of the grid map. To avoid those problems, we use GICP framework instead of using the grid map. In addition, this GICP helps to obtain accurate localization results even though we calibrate motion model parameters roughly.