Two-Dimensional-Simultaneous Localisation and Mapping Study Based on Factor Graph Elimination Optimisation

Abstract

A robust multi-sensor fusion simultaneous localization and mapping (SLAM) algorithm for complex road surfaces is proposed to improve recognition accuracy and reduce system memory occupation, aiming to enhance the computational efficiency of light detection and ranging in complex environments. First, a weighted signed distance function (W-SDF) map-based SLAM method is proposed. It uses a W-SDF map to capture the environment with less accuracy than the raster size but with high localization accuracy. The Levenberg–Marquardt method is used to solve the scan-matching problem in laser SLAM; it effectively alleviates the limitations of the Gaussian–Newton method that may lead to insufficient local accuracy, and reduces localisation errors. Second, ground constraint factors are added to the factor graph, and a multi-sensor fusion localisation algorithm is proposed based on factor graph elimination optimisation. A sliding window is added to the chain factor graph model to retain the historical state information within the window and avoid high-dimensional matrix operations. An elimination algorithm is introduced to transform the factor graph into a Bayesian network to marginalize the historical states and reduce the matrix dimensionality, thereby improving the algorithm localisation accuracy and reducing the memory occupation. Finally, the proposed algorithm is compared and validated with two traditional algorithms based on an unmanned cart. Experiments show that the proposed algorithm reduces memory consumption and improves localisation accuracy compared to the Hector algorithm and Cartographer algorithm, has good performance in terms of accuracy, reliability and computational efficiency in complex pavement environments, and is better utilised in practical environments.