• CN：11-2187/TH
  • ISSN：0577-6686
机械工程学报

机械工程学报 ›› 2023, Vol. 59 ›› Issue (17): 44-55.doi: 10.3901/JME.2023.17.044

• 机器人及机构学 • 上一篇    下一篇

扫码分享

基于强制闭环与隐马尔可夫模型序列匹配的多机器人协同建图

任靖渊1,2, 胡钊政2,3, 陶倩文2, 李娜2,3, 万金杰1,2   

  1. 1. 武汉理工大学信息工程学院 武汉 430070;
    2. 武汉理工大学智能交通系统研究中心 武汉 430063;
    3. 武汉理工大学重庆研究院 重庆 401120
  • 收稿日期:2022-09-05 修回日期:2022-12-23 出版日期:2023-09-05 发布日期:2023-11-16
  • 通讯作者: 胡钊政(通信作者),男,1979年出生,博士,教授,博士研究生导师。主要研究方向为3D计算机视觉理论与应用,视觉与激光SLAM定位,智能交通系统。E-mail:zzhu@whut.edu.cn
  • 作者简介:任靖渊,男,1998年出生。主要研究方向为多机器人激光SLAM,机器人定位。E-mail:renjingyuan@whut.edu.cn;陶倩文,女,1994年出生,博士研究生。主要研究方向为3D计算机视觉,智能车定位技术。E-mail:tqw@whut.edu.cn;李娜,女,1981年出生,硕士。主要研究方向为机器人定位,3D计算机视觉。E-mail:nal926@whut.edu.cn;万金杰,男,1998年出生,硕士研究生。主要研究方向为智能车定位,多传感器融合。E-mail:jinjiewan@whut.edu.cn
  • 基金资助:
    国家重点研发计划（2022YFB2502904）、湖北省重点研发计划（2022BAA082）、武汉市人工智能创新专项（2022010702040064）和武汉理工大学重庆研究院科技创新研发（YF2021-04）资助项目。

Multi-robot Collaborative Mapping Based on Enforced Loop Closure and Sequence Matching with Hidden Markov Model

REN Jingyuan1,2, HU Zhaozheng2,3, TAO Qianwen2, LI Na2,3, WAN Jinjie1,2   

  1. 1. School of Information Engineering, Wuhan University of Technology, Wuhan 430070;
    2. Intelligent Transport System Center, Wuhan University of Technology, Wuhan 430063;
    3. Chongqing Research Institute of Wuhan University of Technology, Chongqing 401120
  • Received:2022-09-05 Revised:2022-12-23 Online:2023-09-05 Published:2023-11-16

PDF

235

摘要： 通过多机器人协作可以高效完成大场景地图构建，但建图质量严重依赖于多机器人间的闭环检测效果。相较于传统的单机器人，多机器人闭环检测问题更为复杂困难。将多机器人闭环检测问题分解为两个独立问题，即闭环存在性问题与闭环存在条件下的多机器人数据匹配问题。针对闭环存在性问题，提出强制闭环策略，即在场景中设定强制闭环区域，强制多机器人访问该区域。在此基础上，提出将多机器人数据匹配问题转化为基于隐马尔可夫模型（Hidden Markov model，HMM）的序列匹配问题。通过对HMM模型中的初始匹配结果、状态转移概率和发射概率进行建模，完成不同机器人采集的数据序列的精准匹配。最后，提出基于大规模位姿图优化（Pose graph optimization，PGO）模型的多机器人协同建图方法，利用序列匹配结果完成约束构建实现多机协同建图。利用标准测试数据集与多机器人实际采集的数据对算法进行验证。实验结果表明，提出的多机协同建图方法不仅有效提升建图效率与自动化程度，同时保证较高的建图质量，算法性能明显优于传统的地图构建方法。

关键词: 大规模场景建图, 多机器人, 隐马尔可夫模型, 位姿图优化

Abstract: Map construction of large scene can be completed efficiently through multi-robot cooperation. However, the quality of map construction deeply depends on loop closure detection results among multiple robots. Compared to single robot case, the multi-robot loop closure detection problem is more complex and difficult. The proposed method decomposes the multi-robot loop closure detection problem into two independent problems, the problem of loop closure existence and the problem of multi-robot data matching under the condition of the loop closure existence. For the problem of loop closure existence, an enforced loop closure strategy is proposed, which is setting an enforced loop closure area in the scene and enforcing multiple robots to access through. On this basis, the data matching problem across multiple robots is formulated into the hidden Markov model (HMM) based sequence matching problem. By modelling the initial matching results, state transition probability, and emission probability in the HMM model, accurate matching of data sequences collected by different robots can be accomplished. Finally, a multi-robot collaborative mapping method based on large-scale pose graph optimization (PGO) model is proposed, which uses sequence matching results to complete constraint construction and realize multi-robot collaborative mapping. The standard test database and the actual field data collected by multiple robots were used to validate the proposed algorithm. Experimental results show that the proposed multi-robot collaborative mapping method not only effectively improves the mapping efficiency and automation, but also ensures high mapping quality. The results also demonstrate the proposed algorithm outperformances traditional mapping methods.

Key words: large scale scene mapping, multi-robot, hidden Markov model, pose graph optimization

中图分类号: 

版权所有 © 2019 《机械工程学报》编辑部 | 京ICP备05055788号-30