CVI-SLAM—Collaborative Visual-Inertial SLAM

Abstract

With robotic perception constituting the biggest impediment before robots are ready for employment in real missions, the promise of more efficient and robust robotic perception in multi-agent, collaborative missions can have a great impact many robotic applications. Employing a ubiquitous and well-established visual-inertial setup onboard each agent, in this paper we propose CVI-SLAM, a novel visual-inertial framework for centralized collaborative SLAM. Sharing all information with a central server, each agent outsources computationally expensive tasks, such as global map optimization to relieve onboard resources and passes on measurements to other participating agents, while running visual-inertial odometry onboard to ensure autonomy throughout the mission. Thoroughly analyzing CVI-SLAM, we attest to its accuracy and the improvements arising from collaboration, and evaluate its scalability in the number of participating agents and applicability in terms of network requirements.

On Data Sharing Strategy for Decentralized Collaborative Visual-Inertial Simultaneous Localization And Mapping

“…Forster et al [2] developed a centralized monocular visual SLAM framework which assigns low-level tasks such as keyframe selection and processing to the agents while high-level tasks for Collaborative Structurefrom-Motion like mapping, loop closing and map merging are performed by a ground station. Karrer et al [3] have recently introduced a similar system for VI-SLAM called CVI-SLAM. However, those methods often require to exchange significant amounts of data that might exceed the available communication bandwidth.…”

On the Redundancy Detection in Keyframe-Based SLAM

“…Building on top of state-of-the-art Simultaneous Localization And Mapping (SLAM) systems with considerable robustness and accuracy in the centimeter range [23] for single-robot scenarios, multi-agent system have been gaining growing popularity in various applications, ranging from inspection tasks to search-and-rescue missions. By sharing information amongst the participants or dividing up a task between multiple robots, robotic teams can increase the robustness, efficiency and accuracy of a robotic mission [16], and enable tasks impossible for a single robot. Stateof-the-art single-agent [22,23] and centralized multi-agent * This research was supported by the Swiss National Science Foundation (SNSF, Agreement no.…”

“…Preprint Version. Accepted July, 2019 [16,25,28] systems generate a global graph (the SLAM graph or map), estimating the trajectory of the participating robots as well as the 3D structure of the environment. With these SLAM systems being able to cover large areas during robotic missions, and the SLAM graph growing constantly during the mission, the graph quickly reaches a size, where operations on it, such as graph optimization, become computationally expensive, therefore affecting the performance of the SLAM system.…”

“…With these SLAM systems being able to cover large areas during robotic missions, and the SLAM graph growing constantly during the mission, the graph quickly reaches a size, where operations on it, such as graph optimization, become computationally expensive, therefore affecting the performance of the SLAM system. The front-end of keyframe-based SLAM runs Visual Odometry (VO) [22,27,28], using only camera feeds, or Visual-Inertial Odometry (VIO) [23,16,25], using camera images and measurements from an Inertial Measurement Unit (IMU), with the goal of estimating the local trajectory and 3D environment (the local map) of the robot. Keyframebased systems choose a subset of the most representative frames from all incoming frames to be stored in the local map, the keyframes (KFs), and are now well-established in the SLAM literature [19,22,25], gaining substantial ground over filtering-based techniques [31].…”

“…While manifesting itself also in single-agent SLAM, this problem of exploding SLAM maps is especially evident in collaborative systems, where multiple agents contribute data simultaneously to the SLAM estimate. As shown in [16] and [28], this issue is currently the main bottleneck for the scalability of centralized collaborative SLAM systems. Likewise, long-term autonomy [6] depends on efficient data management to ensure applicability when continuously feeding experiences of the environment to the system.…”