DiSCo-SLAM: Distributed Scan Context-Enabled Multi-Robot LiDAR SLAM With Two-Stage Global-Local Graph Optimization
DiSCo-SLAM: Distributed Scan Context-Enabled Multi-Robot LiDAR SLAM With Two-Stage Global-Local Graph Optimization
复制标题
DOI：
10.1109/lra.2021.3138156
复制
发表时间：
2022-04-01
影响因子：
5.2
通讯作者：
Englot, Brendan
中科院分区：
文献类型：
--
作者：
Huang, Yewei;Shan, Tixiao;Englot, Brendan
我们提出了一种用于三维激光雷达观测的分布式、多机器人SLAM的新框架。DISCO-SLAM框架是第一个使用轻量级扫描上下文描述符的多机器人SLAM，允许机器人之间高效地交换激光雷达观测数据。此外，我们的框架还包括一个用于分布式多机器人SLAM的两阶段全局和局部优化框架，该框架提供稳定的定位结果，这些结果对未知的初始条件具有弹性，这些初始条件是搜索机器人间环路闭合的典型特征。在不同的多机器人数据集上，我们将我们提出的框架与广泛使用的分布式Gauss-Seidel(DGS)方法进行了比较，定量地证明了它的准确性、稳定性和数据效率。
We propose a novel framework for distributed,multi-robot SLAM intended for use with 3D LiDAR observations. The framework, DiSCo-SLAM, is the first to use the lightweight Scan Context descriptor for multi-robot SLAM, permitting a data-efficient exchange of LiDAR observations among robots. Additionally, our framework includes a two-stage global and local optimization framework for distributed multi-robot SLAM which provides stable localization results that are resilient to the unknown initial conditions that typify the search for inter-robot loop closures. We compare our proposed framework with the widely used distributed Gauss-Seidel (DGS) approach, over a variety of multi-robot datasets, quantitatively demonstrating its accuracy, stability, and data-efficiency.