A novel indoor localization algorithm based on a modified EKF using virtual dynamic point landmarks for 2D grid maps

Localization is a self-position estimation problem and is one of the most critical research areas in autonomous mobile robots. This study presents novel solutions to two significant practical challenges in localization. The initial challenge pertains to the lack of distinct features within the environment, while the second involves the imperfectness of the map. Many studies are based on Extended Kalman Filter (EKF) and they need specific features from the environment for localization. The edges and corners are the popular natural feature types to be used in grid maps. The proposed study promises accurate and fast position tracking in the grid maps which have very few corners and edges. To achieve high performance in such a challenging map, we propose the virtual dynamic point landmark (VDPL) approach in this paper. VDPL does not need specific features such as edges and corners, and any part of the map can be used as a landmark. Since the real-world applications are based on imperfect maps, mostly generated by SLAM (Simultaneous Localization and Mapping) algorithm, the probabilistic distribution of the measurement model and measurement predictions are incorrect. In this study, the position errors that occur as a result of this incorrectness are alleviated by modifying the EKF algorithm. The modified equations of EKF for taking into account the map errors are clearly shown in the paper. The efficiency of the proposed solution is demonstrated in multiple simulations which are performed in randomly generated maps. Moreover, the benefits and real-time performance of the proposed approach are provided by real-world tests using an autonomous wheelchair platform. We believe that the methods developed in this study will improve the localization performance of autonomous robots that operate in challenging environments in terms of feature structure.