A Robust Visual-Inertial SLAM in Complex Indoor Environments

This article proposes a visual-inertial simultaneous localization and mapping (SLAM) method based on edge alignment, which is specifically designed for mobile robot platforms with limited load operating in complex indoor environments, particularly those with weak texture and multiple corners. The proposed algorithm is a modified version of the real-time edge-based SLAM (RESLAM) framework, with the integration of inertial data into both its front- and back-end processing to enhance the positioning robustness of RESLAM. In addition, the algorithm includes inertial initialization and dynamic marginalization processes to ensure the stable operation. The performance of the algorithm has been evaluated through both simulation and experiments. The simulation results indicate that the proposed algorithm reduces the absolute trajectory error (ATE) of positioning error in comparison to the original edge alignment method. The improvement ranges from 13.62% to 71.70%, and it is more pronounced in sequences with complex motion, such as WithMR and Fast. To further validate the effectiveness of the proposed algorithm, a prototype platform was constructed, and its positioning capabilities were verified in indoor environments with weak textures and multiple corners.