LiLoc: Lifelong Localization using Adaptive Submap Joining and Egocentric Factor Graph

TL;DR

LiLoc presents a graph-based lifelong localization framework that maintains a central session and uses subsidiary sessions to achieve long-term, accurate localization. Its core contributions are adaptive submap joining to manage prior submaps, an egocentric factor graph (EFG) that tightly couples IMU, LiDAR odometry, and scan matching via a propagation model, and a mode-switching mechanism that flexibly transitions between relocalization and incremental localization. The approach is validated on public and custom datasets, showing competitive accuracy and robust cross-session performance, with open-source code to follow. Overall, LiLoc advances long-term robotic localization by efficiently reusing central-session knowledge while mitigating uncertainty in cross-session constraints.

Abstract

This paper proposes a versatile graph-based lifelong localization framework, LiLoc, which enhances its timeliness by maintaining a single central session while improves the accuracy through multi-modal factors between the central and subsidiary sessions. First, an adaptive submap joining strategy is employed to generate prior submaps (keyframes and poses) for the central session, and to provide priors for subsidiaries when constraints are needed for robust localization. Next, a coarse-to-fine pose initialization for subsidiary sessions is performed using vertical recognition and ICP refinement in the global coordinate frame. To elevate the accuracy of subsequent localization, we propose an egocentric factor graph (EFG) module that integrates the IMU preintegration, LiDAR odometry and scan match factors in a joint optimization manner. Specifically, the scan match factors are constructed by a novel propagation model that efficiently distributes the prior constrains as edges to the relevant prior pose nodes, weighted by noises based on keyframe registration errors. Additionally, the framework supports flexible switching between two modes: relocalization (RLM) and incremental localization (ILM) based on the proposed overlap-based mechanism to select or update the prior submaps from central session. The proposed LiLoc is tested on public and custom datasets, demonstrating accurate localization performance against state-of-the-art methods. Our codes will be publicly available on https://github.com/Yixin-F/LiLoc.

Figures (5)

• Figure 1: A real-world application of LiLoc with central session A and subsidiary sessions B and C. In the top-right subfigure, the pentagons represent the start points, rectangles denote covered nodes, dashed lines indicate odometry constraints, and solid lines represent prior constraints. Since Session B passes through the area covered by Session A, it switches from ILM to RLM. The bottom-right shows our data collection equipment.
• Figure 2: System overview of our proposed LiLoc, which consists of four main modules: (a) adaptive submap joining, (b) pose initialization, (c) egocentric factor graph (EFG), and (d) mode switching.
• Figure 3: Merged Trajectories resulted from multi-session pose estimation on (a) NCLT dataset and (b) our custom dataset.
• Figure 4: Merged maps from (a) the NCLT dataset and (b) our custom dataset, with the right subfigures showing mapping details to indicate pose estimation accuracy. (a) nclt_120115 (white) is used as the central session, with localization for subsidiary sessions nclt_120429 (green) and nclt_120615 (blue). (b) The same experiment is conducted using central session A (white) and subsidiary sessions B (green) and C (blue).
• Figure 5: Time cost evaluation of two processes: the LM solver (LM Time) and the JFGO using EFG (Opt Time).