A Survey of simultaneous localization and mapping for robot

In unknown environment, it’s very important to obtain pose of the robot and environment map through sensors such as camera and lidar carried by the robot itself. In this paper, several common slam algorithms are introduced, moreover, the advantages and disadvantages of different algorithms are also compared. Traditional slam algorithms mainly include filter-based slam and graph-based slam. The filter-based slam algorithm is mature and easy to understand. It is mainly based on extended Kalman filter (EKF-SLAM) and particle filter (PF-SLAM). However, the linearization error exists in EKF model and the complexity increases sharply when the moving range is large. Although PF-SLAM does not have linearization problem, it cost much to obtain high precision and the algorithm has problem of particle consumption. Slam based on graph optimization can obtain globally consistent pose and map, but the algorithm is ineffective when the environment is in the absence of texture information . The combination of traditional slam and deep learning can improve the robustness of slam algorithm, what’s more, the semantic map generated by deep learning has positive significance for intelligent navigation of robots. With the development of deep learning, sensors and computer technology, slam is bound to make further progress.