SIA OpenIR  > 其他
大范围厂区环境下移动机器人定位与自主导航技术研究
Alternative TitleThe Key Techniques of Localization and Autonomous Navigation for Mobile Robots in Large-Scale Factory Environments
杨奇峰
Department其他
Thesis Advisor曲道奎 ; 徐方
Keyword激光点云 定位 场景理解 路径规划
Pages104页
Degree Discipline机械电子工程
Degree Name博士
2021-08-21
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract移动机器人已广泛应用于工厂的室内环境,完成货物的搬运、装配等工作,其在该类场景下的定位与导航技术已较为成熟。但工业厂区、电站等室外大范围厂区环境,场景动态变化较大,未知性和不确定性较多,给移动机器人自主定位、场景理解和路径规划带来更多挑战。面向大范围厂区的移动机器人自主导航技术,成为移动机器人完成物流运送、安保巡逻等应用的关键技术问题。定位是实现自主导航的基础,工厂厂区环境多样,很多厂区存在建筑物、树木遮挡,强电磁干扰(如变电站)等影响定位的不利因素,实现鲁棒性高且精度高的定位技术对移动机器人室外应用至关重要;大范围厂区环境的多样性和高动态性给移动机器人场景理解的识别准确度和实时性带来严峻挑战,也对机器人安全高效的自主运行构成威胁;同时行人、车辆等动态障碍物对移动机器人的动态避障路径规划技术提出了更高的要求。本文围绕面向大范围厂区环境移动机器人的定位与自主导航关键技术问题展开研究,主要研究内容与创新点如下:1、为解决室外大范围厂区环境下移动机器人高精度定位的鲁棒性问题,提出采用二维激光旋转获取三维稠密点云,构建先验拓扑-点云地图,在此基础上基于3D-NDT三维激光点云实时匹配,融合轮式里程计数据实现移动机器人室外环境下的精确定位,解决了移动机器人在不依赖GPS定位信息的情况下完成在室外大范围厂区环境中的快速精确定位问题。实验结果表明,该方法具有精度高、鲁棒性与实时性好的特点。2、为提高移动机器人室外环境场景理解的识别准确度和实时性,提出基于激光点云几何特征和点云减法密度聚类的典型障碍物识别算法,解决园区中马路路牙和行人的快速识别问题。同时针对基于视觉的场景理解方法对室外环境中光照动态变化适应性差,而利用激光雷达实现环境感知,直接在三维空间中进行目标动态识别,数据量大、耗时长的问题,提出将三维点云转换为二维图像(BA图),采用掩模区域卷积神经网络(Mask R-CNN)实现场景理解,在此基础上利用BA图像素与激光点云中激光点一一对应的特点,实现BA图中所识别物体在三维点云中的快速定位。实验结果证实该算法能满足移动机器人实时应用的需求,静态与动态障碍物的准确快速识别为提高机器人路径规划的安全性奠定了基础。3、针对移动机器人在室外厂区动态变化的环境中,避障和路径规划的安全性问题,本文构建了拓扑地图进行路网建模,并以此为基础实现全局路径规划,同时提出基于障碍物运动趋势预测的动态路径规划方法:建立了依托障碍物历史运动轨迹的动态障碍物运动预测模型,在移动机器人的动态避障与路径规划过程中,不仅考虑障碍物当前的位置,同时评估动态障碍物的移动轨迹,在此基础上提出改进的 D*Lite 局部路径规划算法,使机器人的动态避障路径规划算法更接近于人的避障策略。实验验证证明所提方法相比 D*Lite 算法,显著提高了移动机器人避障路径规划的效率和安全性。4、在解决上述关键技术的基础上,搭建了移动机器人实验平台,设计了基于李雅普诺夫级联控制的机器人运动控制器,实现了移动机器人良好的动态响应性能和对规划路径的高精度跟踪,完成了移动机器人在室外大范围厂区环境下的定位、场景理解与避障路径规划算法验证。实验结果显示所提算法可以使移动机器人在室外厂区动态环境下实现精确定位和自主导航运行,为今后移动机器人面向室外厂区安保、巡检、物流等应用和产业化发展奠定了基础。
Other AbstractMobile robot has been widely used in factories for indoor tasks such as transportation and assembly. The positioning and navigation techniques of mobile robot are relatively mature. but in the large-scale outdoor environment such as industrial area and power stations, the scene changes greatly, has many unknown and uncertain factors. These factors bring more challenges to the autonomous positioning, scene understanding and path planning of mobile robots. The autonomous navigation technology for large-scale industrial area has become one of the key technical problems in logistics transportation, security patrol and other applications. Positioning is the basis of autonomous navigation. There are many adverse factors such as buildings, trees, strong electromagnetic interference (such as substation) in many factory areas. It is important to realize the positioning technology with high robustness and high precision for the outdoor applications of mobile robots. Meanwhile, the diversity and high dynamics of the large-scale plant environment bring serious challenges for the mobile robot to the recognition accuracy and real-time performance of the scene understanding, and also pose a threat to the safe and efficient autonomous operation of the robot. In addition, the dynamic obstacles such as pedestrians and vehicles put forward higher requirements for the dynamic obstacle avoidance path planning technology of mobile robots. This paper focuses on the key technology of localization and autonomous navigation of mobile robot for large-scale industrial environment. 1. To solve the robustness problem of high-precision positioning of mobile robot in the outdoor large-scale factory environment, a two-dimensional laser with a rotation axis is designed to obtain three-dimensional dense point cloud, and a priori topology-point-cloud map is constructed. Then, based on 3D-NDT laser point cloud, real-time matching and wheel odometer data, the accurate positioning of mobile robot in outdoor environment is realized. This method can realize fast and accurate positioning in the outdoor large-scale factory environment without relying on GPS information. The experimental results show this method has the characteristics of high precision, good robustness and real-time performance. 2. To improve the recognition accuracy and real-time performance of scene understanding for mobile robot in outdoor environment, a typical obstacle recognition algorithm based on geometric characteristics of laser point cloud is proposed to solve the problem of rapid identification of road and pedestrian in the plant. Meanwhile, to solve the problem of poor adaptability of vision based scene understanding method in outdoor environment, and using lidar takes a long time and poor real-time performance to recognize targets in three-dimensional space directly. The 3D point cloud is transformed into 2D image (BA map), and Mask R-CNN is used to realize scene understanding. Then, the fast location of the object recognized in the 3D point cloud is realized by using the feature that the pixels of BA image correspond to the laser points in the laser point cloud one by one. The experimental results show that the algorithm can meet the needs of real-time application of mobile robot, and the accurate and rapid identification of static and dynamic obstacles lays a foundation for improving the safety of robot path planning. 3. In view of the safety of obstacle avoidance and path planning for mobile robot in the dynamic environment of outdoor plant, the global path planning is realized based on constructs a topological map for road network. Meanwhile, a dynamic path planning method based on the motion prediction of obstacles is proposed: the dynamic obstacle motion prediction based on the historical trajectory of obstacles is established in the process of dynamic obstacle avoidance and path planning of mobile robot. Not only the current position of obstacles is considered, but also the moving trajectory of dynamic obstacles is evaluated. On this basis, an improved D* Lite local path planning algorithm is proposed, which makes the dynamic obstacle avoidance path planning algorithm of robot closer to the human obstacle avoidance strategy. Experimental results show that the proposed method significantly improves the efficiency and security of obstacle avoidance path planning of mobile robot compared with D* Lite algorithm. 4. Based on the aforementioned key techniques, a mobile robot experimental platform is built, and then a cascaded robot motion controller is designed based on the Lyapunov theory. The proposed method has good dynamic response performance and high-precision tracking of the planned path. The positioning, scene understanding and obstacle avoidance path planning algorithms are verified in the outdoor large-scale factory environment. The experimental results show that the proposed algorithm can achieve accurate positioning and autonomous navigation in the dynamic environment of the outdoor plant. The proposed method establishes a foundation for the industrial applications and developments of future mobile robot in outdoor plant security, inspection, and logistics.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/29404
Collection其他
Affiliation中国科学院沈阳自动化研究所
Recommended Citation
GB/T 7714
杨奇峰. 大范围厂区环境下移动机器人定位与自主导航技术研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2021.
Files in This Item:
File Name/Size DocType Version Access License
大范围厂区环境下移动机器人定位与自主导航(5129KB)学位论文 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[杨奇峰]'s Articles
Baidu academic
Similar articles in Baidu academic
[杨奇峰]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[杨奇峰]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.