Design And Implementation Of Simultaneous Localization And Mapping System For Undulating Road Environments

In recent years,the research on domestic unmanned driving has risen to the national strategic level.Simultaneous localization and mapping(SLAM),as a core technology in the field of unmanned driving,has received significant attention and become the core and focus of research.SLAM can build 3D maps in unknown environments with no prior information.It can generate high-precision,comprehensive 3D point cloud maps that showcase environmental information,even in highly uncertain and complex terrains.SLAM provides map support for autonomous localization and navigation of mobile robots,enabling them to played a significant role in high-risk and time-sensitive missions such as forest fire fighting and post-earthquake rescue.SLAM relies on external sensors to gather information about the environment,thereby estimating the vehicle’s pose and completing a map.Light detection and ranging(Li DAR)is unaffected by environmental factors and can accurately obtain the target position.Therefore,it is widely used in SLAM algorithms.At present,Li DAR-based SLAM algorithms are mostly applied to structured and flat urban road scenarios.However,in practical SLAM application scenarios,there are also non-structured pavement environments,and the terrain is uncertain,such as off-road,mountains,and Gobi.The existence of these scenes makes the commonly used SLAM methods based on the assumption of planar motion no longer applicable.Thus,this paper proposes a Li DARbased SLAM algorithm for undulating road environments,the specific research contents are as follows:Firstly,this paper describes a feature point extraction method that addresses the challenge of insufficient extraction of valuable features due to small overlap areas in non-structured(undulating road)conditions.Specifically,the method uses the distribution of neighborhood normal vector angles to obtain points with surface information,and then leverages the local surface characteristics of the point cloud to extract discriminative features,thereby enhancing the ability of feature points to describe the undulating road surface.Secondly,in undulating road scenarios,the vehicle shakes due to bumps,causing the point clouds have pitch.Applying traditional registration methods based on distance constraints have low accuracy.Therefore,this paper improves the Iterative Closest Point(ICP)registration algorithm by introducing a multi-scale constraint description based on point cloud curvature,normal vector angle,and Euclidean distance.This enhances the discriminative power of the algorithm between feature points and helps prevent mis-matching.At the same time,a dynamic iteration factor is introduced in the registration process,and the correspondence between matching points is corrected by adjusting the distance and angle threshold to reduce the impact of pose initialization and improve the robustness of point cloud registration.Furthermore,relying only on the Euclidean distance between two frames under the situation of point clouds with pitch cannot ensure the stability of loop detection,resulting in missed detection.Therefore,this paper proposes a loop closure detection method based on surface curvature-Euclidean distance bidirectional compensation,which finds loop closures through local neighborhood surface features,and then uses distance constraints to supplement the known loop frames,improving the method’s ability to recognize loop closure frames in scenes with undulating transformations.Finally,this paper establishes an experimental platform and conducts experiments using Campus and Off-road datasets to demonstrate the excellent performance of the proposed algorithm in registering and mapping undulating road terrains.