LT-mapper: A Modular Framework for LiDAR-based Lifelong Mapping

Giseop Kim, Ayoung Kim

Abstract

Long-term 3D map management is a fundamental capability required by a robot to reliably navigate in the non-stationary real-world. This paper develops open-source, modular, and readily available LiDAR-based lifelong mapping for urban sites. This is achieved by dividing the problem into successive subproblems: multi-session SLAM (MSS), high/low dynamic change detection, and positive/negative change management. The proposed method leverages MSS and handles potential trajectory error; thus, good initial alignment is not required for change detection. Our change management scheme preserves efficacy in both memory and computation costs, providing automatic object segregation from a large-scale point cloud map. We verify the framework's reliability and applicability even under permanent year-level variation, through extensive real-world experiments with multiple temporal gaps (from day to year).

Introduction

Autonomous robots operating in real-world urban environments face the significant challenge of non-stationarity. Over time, buildings are constructed, vegetation grows or is removed, and vehicles move. To maintain reliable navigation, a robot's internal map must evolve alongside these changes. Existing lifelong mapping solutions often struggle with computational scalability, sensitivity to alignment errors between sessions, and the unsustainable storage requirements of full 3D point cloud snapshots.

LT-mapper addresses these challenges through a modular pipeline designed for efficiency and robustness. It handles "permanent" year-level variations while minimizing storage overhead by up to 60%.

The LT-mapper Framework

The system is divided into three primary modules that communicate via a file-based protocol, ensuring flexibility and independent operation.

LT-SLAM: Multi-Session Alignment

LT-SLAM serves as the framework's "stitching" engine. It utilizes a multi-session SLAM (MSS) approach with anchor nodes to align data from temporally distant sessions. By employing Scan Context for robust loop detection, the system jointly optimizes trajectories into a shared coordinate system, eliminating the need for perfect initial alignment.

LT-removert: Robust Change Detection

This module distinguishes between High Dynamic (HD) objects (e.g., pedestrians) and Low Dynamic (LD) changes (e.g., new construction). It utilizes a "remove-then-revert" algorithm based on range-image visibility, which maintains accuracy even in the presence of minor SLAM alignment errors.

LT-map: Delta Map Management

Instead of storing redundant full maps, LT-map introduces delta maps. These store only the additions (positive differences) and deletions (negative differences) between sessions. This approach allows for "change composition," where changes between arbitrary sessions can be derived mathematically, enabling efficient map rollbacks and the generation of a stable "Meta Map."

The modular pipeline of the LT-mapper system, illustrating the flow from raw sensor data to managed delta maps.

Experimental Evaluation

The framework was validated using the MulRan Dataset (KAIST sequences) and the newly introduced LT-ParkingLot Dataset. The KAIST sequences provided a rigorous test case with a 1.5-year temporal gap, featuring significant structural developments.

Key Results

  • Storage Efficiency: Delta maps significantly reduce data requirements. In one instance, a delta map required only 0.47 million points compared to 11 million for the full map—a 95.7% reduction.
  • Computational Performance: The change management scheme was found to be 8.9 times faster than traditional methods for computing differences between arbitrary sessions.
  • Structural Robustness: LT-mapper successfully identified permanent structural changes, such as new construction walls, across year-long intervals.
Visualization of structural changes over a 1.5-year gap in the KAIST dataset, showing detected new structures.

Conclusion

LT-mapper provides a scalable, open-source solution for LiDAR-based lifelong mapping. By leveraging modularity and the concept of delta maps, it achieves high storage efficiency and robustness to environmental change. Future work will explore the integration of semantic information and multi-modal sensor data to further enhance the distinction between seasonal and permanent structural variations.

Summary updated

Abstract

Long-term 3D map management is a fundamental capability required by a robot to reliably navigate in the non-stationary real-world. This paper develops open-source, modular, and readily available LiDAR-based lifelong mapping for urban sites. This is achieved by dividing the problem into successive subproblems: multi-session SLAM (MSS), high/low dynamic change detection, and positive/negative change management. The proposed method leverages MSS and handles potential trajectory error; thus, good initial alignment is not required for change detection. Our change management scheme preserves efficacy in both memory and computation costs, providing automatic object segregation from a large-scale point cloud map. We verify the framework's reliability and applicability even under permanent year-level variation, through extensive real-world experiments with multiple temporal gaps (from day to year).
BibTeX
@inproceedings{kim2022lt,
  author = {Giseop Kim and Ayoung Kim},
  title = {{LT-mapper: A Modular Framework for LiDAR-based Lifelong Mapping}},
  booktitle = {IEEE International Conference on Robotics and Automation},
  year = {2022},
  url = {https://openreview.net/forum?id=hx0KmpLGCo},
}
Figure