EdgeSLAM 1.0: Architectural Innovations in Mobile Visual SLAM

In this chapter, we aim to exhibit the system architecture of edge-assisted mobile visual SLAM—that is, how visual SLAM can be accelerated with edge computing. Location services, critical for a myriad of applications, have garnered extensive attention from both the academic sphere and the industry sector. The rise of visual SLAM in domains such as robotics and autonomous vehicles marks a significant advancement. Nonetheless, its deployment on devices with limited resources is hampered by increasing computational demands. This chapter details the development, deployment, and assessment of edgeSLAM (a.k.a., EdgeSLAM 1.0), a cutting-edge service enabling semantic visual SLAM in real-time on mobile platforms. edgeSLAM capitalizes on advanced semantic segmentation techniques to refine both localization and mapping precision and mitigates the processing burden of demanding SLAM and segmentation tasks through strategic computation offloading. The pivotal breakthroughs of edgeSLAM encompass a streamlined offloading approach, a strategic data exchange protocol, and a dynamic task allocation framework. Comprehensive implementation on an edge framework alongside diverse mobile units (including two smartphone models and a developer board) supports edgeSLAM’s efficacy. Rigorous testing across three datasets reveals that edgeSLAM sustains operation at a 35FPS rate while reaching up to 5 cm in localization accuracy, surpassing existing methodologies by over 15%. Furthermore, edgeSLAM’s applicability is validated in pedestrian tracking and robotic navigation scenarios. To our knowledge, edgeSLAM stands as the pioneering endeavor to facilitate real-time semantic visual SLAM on mobile apparatus. We fully implement edgeSLAM on an edge server and different types of mobile devices (2 types of smartphones and a development board). Extensive experiments are conducted under 3 datasets, and the results show that edgeSLAM is able to run on mobile devices at 35FPS frame rate and achieves a 5 cm localization accuracy, outperforming existing solutions by more than 15%. We also demonstrate the usability of edgeSLAM through 2 case studies of pedestrian localization and robot navigation. To the best of our knowledge, edgeSLAM is the first real-time semantic visual SLAM for mobile devices.