MeSLAM: Memory Efficient SLAM based on Neural Fields | IEEE Conference Publication | IEEE Xplore

MeSLAM: Memory Efficient SLAM based on Neural Fields

Publisher: IEEE

Abstract:

Existing Simultaneous Localization and Mapping (SLAM) approaches are limited in their scalability due to growing map size in long-term robot operation. Moreover, processi...View more

Abstract:

Existing Simultaneous Localization and Mapping (SLAM) approaches are limited in their scalability due to growing map size in long-term robot operation. Moreover, processing such maps for localization and planning tasks leads to the increased computational resources required onboard. To address the problem of memory consumption in long-term operation, we develop a novel real-time SLAM algorithm, MeSLAM, that is based on neural field implicit map representation. It combines the proposed global mapping strategy, including neural networks distribution and region tracking, with an external odometry system. As a result, the algorithm is able to efficiently train multiple networks representing different map regions and track poses accurately in large-scale environments. Experimental results show that the accuracy of the proposed approach is comparable to the state-of-the-art methods (on average, 6.6 cm on TUM RGB-D sequences) and outperforms the baseline, iMAP*. Moreover, the proposed SLAM approach provides the most compact-sized maps without details distortion (1.9 MB to store 57 m 3 ) among the state-of-the-art SLAM approaches.
Date of Conference: 09-12 October 2022
Date Added to IEEE Xplore: 18 November 2022
ISBN Information:

ISSN Information:

Publisher: IEEE
Conference Location: Prague, Czech Republic

I. Introduction

Nowadays, the area of autonomous robotics is developing at a very high pace. Its market constituted 1,61 billion USD in 2021 and is expected to grow 13 times by 2030 [1]. For the last decade, mobile robots have been successfully implemented in many areas including goods delivery [2], warehouse logistic [3] –[6], autonomous transport, disinfection [7], [8], and agriculture [9]. Autonomous robots are starting to work alongside humans executing tasks of increasing complexity. Such wide area of application requires robotic systems, including both hardware and software parts, to be robust, safe and efficient in challenging environments, e.g. in day and night conditions [10]. As a part of the software architecture, autonomous robots typically have a perception subsystem that mainly solves the problem of Simultaneous Localization and Mapping (SLAM). It generates a map and defines the position of the robot within the map at the same time. Thus, SLAM is a crucial task that must be solved accurately and efficiently to perform primary robot operation. Currently, many research collectives, both academic and industrial, are aimed at developing new SLAM approaches, designing task-driven methods and improving the existing pipelines to increase their robustness.

Distribution of regions assigned to a single neural network for rendering. Top: in green – left region, in blue – right region, in between – regions intersection. In the middle: corresponding regions rendering. At the bottom: the final merged global map without visible stitching.

References

References is not available for this document.