maplab: An Open Framework for Research in Visual-inertial Mapping and Localization (1711.10250v1)

Abstract: Robust and accurate visual-inertial estimation is crucial to many of today's challenges in robotics. Being able to localize against a prior map and obtain accurate and driftfree pose estimates can push the applicability of such systems even further. Most of the currently available solutions, however, either focus on a single session use-case, lack localization capabilities or an end-to-end pipeline. We believe that only a complete system, combining state-of-the-art algorithms, scalable multi-session mapping tools, and a flexible user interface, can become an efficient research platform. We therefore present maplab, an open, research-oriented visual-inertial mapping framework for processing and manipulating multi-session maps, written in C++. On the one hand, maplab can be seen as a ready-to-use visual-inertial mapping and localization system. On the other hand, maplab provides the research community with a collection of multisession mapping tools that include map merging, visual-inertial batch optimization, and loop closure. Furthermore, it includes an online frontend that can create visual-inertial maps and also track a global drift-free pose within a localization map. In this paper, we present the system architecture, five use-cases, and evaluations of the system on public datasets. The source code of maplab is freely available for the benefit of the robotics research community.

• The paper introduces maplab as a comprehensive, modular framework that integrates online (ROVIOLI) and offline mapping capabilities for visual-inertial localization.
• It employs state-of-the-art visual-inertial odometry and multi-session map merging to achieve accurate pose estimation and drift correction.
• Its extensible, ROS-based, plugin architecture enables rapid experimentation and deployment across diverse robotic platforms and scenarios.

Overview of "maplab: An Open Framework for Research in Visual-inertial Mapping and Localization"

The paper introduces maplab, an open, research-oriented framework designed to advance visual-inertial mapping and localization. The significance of robust visual-inertial estimation systems in robotic applications is well acknowledged, given their importance in addressing challenges related to precise localization and pose estimation. However, existing solutions tend to lack a complete and scalable multi-session mapping capability or an end-to-end pipeline necessary for comprehensive research in this domain.

Maplab stands out as a comprehensive system that integrates state-of-the-art algorithms and provides end-to-end capabilities. It is engineered to support the creation, deployment, and evaluation of multi-session mapping algorithms. The design of maplab places clear emphasis on modularity and extensibility, governed by a C++ implementation structured within the ROS framework using a plugin-based architecture.

System Architecture and Features

Maplab consists of two main components: the online frontend, ROVIOLI (ROVIO with Localization Integration), and the offline maplab console. ROVIOLI allows for real-time visual-inertial odometry by incorporating observations from multiple map sessions, thereby facilitating robust localization and drift correction via mapping. The maplab console, on the other hand, serves as an extensive toolkit for offline map processing, offering map merging, loop closure, visual-inertial optimization, and data sparsification.

A notable aspect of maplab is its ability to handle multi-session mapping tasks efficiently. This capability is crucial in real-world scenarios where mapping needs span multiple sessions over varied conditions. The architecture of maplab is focused on flexibility, allowing for the seamless integration and experimentation with novel mapping algorithms and the rapid deployment across diverse robotic platforms.

Experimental Evaluation and Use Cases

The framework has been evaluated against standard benchmarks using public datasets like EuRoC, demonstrating its competence in delivering accurate pose estimates. The paper presents extensive use cases ranging from online mapping and localization to large-scale and dense mapping. Maplab's operations have been tested over large university environments showcasing its capability to handle complex multi-session and multi-platform scenarios effectively.

Maplab's map maintenance capabilities are pivotal for operations demanding high performance without exhaustive data storage, such as reducing the landmark count while maintaining localization accuracy. This is achieved using landmark summarization techniques, which optimize map size without significant sacrifices in performance.

Implications and Future Research Directions

Maplab offers a unified and extensible platform for visual-inertial mapping and localization research, enabling the advancement of SLAM systems. The extensibility of maplab facilitates the development and integration of new visual-inertial algorithms, potentially directed towards improvement in robustness and efficiency under varying conditions like significant environmental changes.

The implications of maplab extend to its applicability in diverse environments and use-cases including autonomous navigation, robotic manipulation, and augmented reality. Future research might focus on enhancing the system's adaptability to dynamic environments, improving real-time performance, and further integrating various sensors and data types to create richer maps.

Given maplab's open-source nature, it is well-positioned to serve as a foundational platform for academic research and industry applications, encouraging collaborative development and innovation in the field of robotic mapping and localization.