Mobile Robot Localization Through Online SLAM with Modifications.

For autonomous navigation, we have thus far proposed MCL using environmental maps based on cadastral data. However, buildings in the cadastral data sometimes differ from their actual positions in the environment. As the environmental map is generated from the cadastral data, the inconsistency affects the localization performance. For this problem, we propose an online SLAM approach in the actual environment. A mobile robot simultaneously localizes the position and builds another online map using NDT scan matching. In contrast to other offline SLAM approaches, however, pose graph optimization for loop closure is not executed during online SLAM. As a result, the online map is distorted by localization errors. For this challenge inherent in online SLAM, the localization errors are modified using MCL and wheel odometry in a hybrid manner. As a contribution to autonomous navigation, the robot is enabled to localize the position even in a new place. In the experiments, we show that the localization performance of the robot in an outdoor environment with inconsistent buildings is improved compared to other online approaches with and without modifications.