Since the traditional algorithm will cause great deviation when perform large-scale simultaneous mapping and localization.A two-step loop closure detection method based on normal distribution transformation(NDT) is proposed in this paper,which makes full use of NDT to register the midpoint cloud mean and variance characteristics,and adds the proposed method to the SLAM complete framework.Point cloud registration using overlapping grids.Firstly,a point cloud appearance description based on each grid feature value to implement coarse loop closure detection is constructed.If the coarse loop closure condition is met,in order to make the point cloud have rotation invariance,the distance between point cloud grid mean and coordinate origin is calculated to realize accurate loop detection.The proposed algorithm is verified on “small cyclone” autonomous car platform.Experiments show that the algorithm can effectively reduce the cumulative deviation in large-scale construction,the robustness of the system becomes stronger,and the tracking performance becomes better.