In order to improve the accuracy of point cloud reconstruction for automatic drive sweeping robots, a simultaneous localization and mapping (SALM) algorithm based on graph optimization is proposed. First, the extended Kalman filter is used to fuse the information of GPS, inertial measurement unit (IMU) and odometer to get the current position. Second, the point cloud transformation relationship is obtained based on 3D-NDT registration. Finally, by constructing point clouds as map nodes, GPS and ground parameters as edge constraints, the back-end optimization is carried out by constructing a map optimization model. The point cloud posture is constructed as a map node, and the real-time laser point cloud data, fusion location information and ground parameters are used as edge constraints, and solve the optimum position and posture of point clouds. The results show that comparing with mapping algorithms that just based on laser data, the proposed algorithm can improve the mapping results of point cloud environment and improve the mapping accuracy. The correctness and efficiency of the strategy in this paper is verified.