Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Efficient Localization for Multi-Agent SLAM
The goal of this project is to improve the localization capabilities of the visual-inertial odometry front-end of collaborative SLAM systems, actively switching between localization and mapping depending on the available information about the environment.
Keywords: SLAM, Visual Odometry, Localization, Multi-Robot Systems
State-of-the-art Simultaneous Localization And Mapping (SLAM) systems have reached a high level of maturity in single-agent [1] and multi-agent [2] scenarios. In addition to reaching a higher level of accuracy compared to single-agent SLAM, [2] has shown that the re-use of shared data in multi-agent SLAM can improve the accuracy of the Visual-Inertial Odometry (VIO) front-end of the collaborative SLAM system. However, this information re-use is still very limited in contemporary multi-agent SLAM approaches: even when the area the robot is moving in is already exhaustively mapped, only a small portion of this map information can be digested by the VIO front-end.
The goal of this project is to investigate approaches to improve the re-use of shared data by the VIO front-end in a collaborative SLAM setup. Especially, the approach should dynamically adapt to the map information present in the currently explored environment: if the scene is sufficiently mapped, the VIO should fall back to localizing against the existing map, while switching back to actively mapping the environment as soon as the well-mapped area is left. Such a strategy has the potential to significantly improve the robustness and efficiency of contemporary collaborative SLAM systems, boosting the scalability and real-world applicability of this technology.
References:
[1] Qin et al.: VINS-Mono: A Robust and Versatile Monocular Visual-Inertial State Estimator, T-RO 2018
[2] Karrer et al.: CVI-SLAM – Collaborative Visual-Inertial SLAM, RA-L 2018
State-of-the-art Simultaneous Localization And Mapping (SLAM) systems have reached a high level of maturity in single-agent [1] and multi-agent [2] scenarios. In addition to reaching a higher level of accuracy compared to single-agent SLAM, [2] has shown that the re-use of shared data in multi-agent SLAM can improve the accuracy of the Visual-Inertial Odometry (VIO) front-end of the collaborative SLAM system. However, this information re-use is still very limited in contemporary multi-agent SLAM approaches: even when the area the robot is moving in is already exhaustively mapped, only a small portion of this map information can be digested by the VIO front-end.
The goal of this project is to investigate approaches to improve the re-use of shared data by the VIO front-end in a collaborative SLAM setup. Especially, the approach should dynamically adapt to the map information present in the currently explored environment: if the scene is sufficiently mapped, the VIO should fall back to localizing against the existing map, while switching back to actively mapping the environment as soon as the well-mapped area is left. Such a strategy has the potential to significantly improve the robustness and efficiency of contemporary collaborative SLAM systems, boosting the scalability and real-world applicability of this technology.
References:
[1] Qin et al.: VINS-Mono: A Robust and Versatile Monocular Visual-Inertial State Estimator, T-RO 2018