结合前文的理论，设计了多传感器图优化SLAM系统，包括四个模块，如图4

Combined with the previous theory, a multi-sensor graph optimized SLAM system is designed, including four modules, as shown in Figure 43. The data acquisition module collects sensor data, processes it, and sends it to the pose acquisition module. The IMU and the encoder are fused to generate odometer information. The 2D laser and RGB-D point cloud are scanned and matched to synthesize the pose, and then the UKF is fused to improve the positioning accuracy. The RGB-D point cloud is also used to solve the descriptor and simplify The point cloud information is optimized in the back-end optimization module through loopback detection to restore the actual movement posture of the robot. The map creation module creates a raster map based on the 2D point cloud generated by the 2D laser point cloud and the RGB-D point cloud dimensionality reduction.

Combined with the previous theory, a multi-sensor diagram is designed to optimize the SLAM system, including four modules, as shown in Figure 43. The data acquisition module collects the sensor data, processes it, and sends it to get the module. IMU and encoder fusion, generate odometer information, 2D laser and RGB-D point cloud are scanned to match the synthetic position, and then through UKF fusion, improve positioning accuracy, RGB-D point cloud at the same time to describe sub-solution, simplify point cloud information, optimize the module in the back end, optimize through loop detection, restore the actual movement position of the robot. The map creation module creates a grid map based on 2D point clouds generated by 2D laser point clouds and RGB-D point cloud downdimension.

Combined with the theory of the previous paper, a multi-sensor graph optimization slam system is designed, including four modules, as shown in Figure 43. The data acquisition module collects the sensor data, processes it, and sends it to the pose acquisition module. IMU and encoder are fused to generate odometer information. 2D laser and rgb-d point cloud are scanned and matched to synthesize pose respectively. Then, UKF is used to fuse to improve positioning accuracy. Rgb-d point cloud is solved at the same time to simplify point cloud information. In the back-end optimization module, loop detection is used to optimize and recover the actual motion pose of the robot. The map creation module creates raster map based on 2D laser point cloud and rgb-d point cloud.<br>