Mapping And Localization Technology For Mobile Robot In Large-Scale Outdoor Environments

In recent years,the mobile robot has more and more applications and developments in the field of life services,factory logistics,disaster relief,etc.Simultaneous Localization and Mapping(SLAM)is the most basic and core technology.When mobile robots explore unknown environments,SLAM technology can be used to construct environmental maps,which is the basic task for mobile robots to complete subsequent tasks.The localization technology based on map is the basic link to realize the autonomy of mobile robots,and it is also the premise of completing tasks such as autonomous navigation and inspection.It can be said that the mapping and localization technology is the most important for the development of mobile robots.Although it has been applied matured in indoor structured scenes,there are still many difficulties and challenges in outdoor large-scale scenes,such as roads,farmlands,and electrical substations.Due to the lack of features in these scene,the unreasonable data association may destroy the consistency of the map,the largescale scene may causes the global localization cannot be quickly converged,or the scene similarity brings about a problem that the localization convergence accuracy is not high.This paper focuses on mobile robots equipped with 3D Li DAR sensor operating in outdoor large-scale environments,and conducts detailed analysis and research on key technologies for mapping and localization,in order to improve the accuracy of mapping and the convergence speed and robustness of localization without the Global Positioning System(GPS)in the scene with the characteristics of sparse,similarity or high dynamics.The research content of this paper is as follows:1.Aiming at the challenges of mapping technology,incremental optimization SLAM method based on multi-layer ICP matching is proposed in this paper.Specifically,for a large amount of dynamic and noise information in the outdoor environment may cause the increase of uncertainty of the matching algorithm,and a reliable point cloud down sampling strategy is proposed,which can effec-tively extract static and stable point cloud data;In order to ensure the real-time performance of SLAM and the robustness of data association when there are lack of features in the scene,a multi-layer ICP matching algorithm is proposed,which combines Scan-to-Scan ICP and Scan-to-Sub Map ICP matching to construct real-time low drift Li DAR odometer,and Scan-to-Key Scan ICP matching is used to solve the data association problem when the robot returns to the visited area to avoid the cumulative error in a wide range of scenarios,which may damage the consistency of the global map.The SLAM problem is finally reduced to a least squares problem and solved using an incremental optimization algorithm,which can improve the calculation efficiency.The proposed mapping algorithm can achieve high-precision mapping and meet real-time applications.2.Localization technology can be divided into two parts: global localization and pose tracking.It is a very challenging task to realize fast global localization in large-scale scenes without relying on GPS.This paper proposes a novel localization framework.It is divided into two parts: place recognition and pose estimation,which can improve the convergence speed and robustness of the localization under the condition of large-scale environment and high similarity of the scene.Specifically,it is difficult for the traditional particle filter algorithm to quickly converge to the initial pose in a wide range of scenarios.Therefore,a place recognition algorithm based on random forest learning can be used to quickly search the initial pose.In order to avoid the traditional particle filtering algorithm converging to the wrong single and high likelihood region in the high similarity scenario,the multi-hypothesis particle filtering algorithm is proposed to independently develop and track multiple initial pose hypothesis sets until converge to the correct pose hypothesis.Finally,for the large-scale environment and the scene has a high similarity,the proposed algorithm can improve the convergence speed,accuracy and robustness.In summary,this paper proposes a solution for solving the mobile robots mapping and localizing in large-scale outdoor scenes.We verified the proposed algorithms in multiple mobile platforms,and a series of real scenes and public data sets.Experiments prove the effectiveness of the proposed algorithm and its superiority in accuracy and time.