SERail-SLAM: Semantic-Enhanced Railway LiDAR SLAM

Weiwei Song, Shiqi Zheng, Xinye Dai, X. Wang, Y. N. Wang, Z. Y. Wang, Shujie Zhou, Wei Liu, Yidong LouMachines 2026 · DOI

TL;DR

This paper introduces SERail-SLAM, a novel LiDAR SLAM system specifically designed for the challenging railway environment. By integrating semantic information with geometric data and employing adaptive mapping and filtering techniques, SERail-SLAM significantly improves localization accuracy and robustness, especially in GNSS-denied or feature-scarce railway corridors.

What Problem Does This Paper Address?

Autonomous railway operations, like maintenance and navigation, demand highly precise and reliable localization. However, railway environments pose unique challenges for traditional SLAM systems. The repetitive linear structures lead to geometric degeneracy, making it difficult to distinguish between frames. High-speed train operation causes point cloud sparsity, and rigid downsampling methods can discard crucial features. Furthermore, the mix of static and dynamic objects, along with environmental noise, can lead to inaccurate mapping and drift. Existing semantic SLAM methods often rely on discrete labels and struggle with segmentation uncertainty and varying object stability, failing to fully leverage semantic cues for robust localization in these complex scenarios.

Methods

SERail-SLAM tackles these challenges with a multi-pronged approach:

  • Semantic-Enhanced Adaptive Voxel Map: Instead of fixed-resolution voxel grids, this method uses eigen-decomposition of local point distributions to create an adaptive map. This allows for preserving fine-grained structures like rails while compressing redundant planar surfaces, leading to a more descriptive and efficient map.
  • Confidence-Aware Filtering: This technique goes beyond simple semantic masking. It uses the raw probability outputs from semantic segmentation networks to adaptively weight input measurements. This allows the system to distinguish reliable landmarks from noise and clutter, especially in areas with uncertain segmentation.
  • Category-Weighted Joint Optimization: The system incorporates semantic stability priors into the optimization process. Features are weighted based on the structural reliability of their semantic class, ensuring that stable features (like poles and rails) have a stronger influence on localization, mitigating longitudinal drift.
  • Tightly Coupled Multi-Sensor Fusion: SERail-SLAM integrates LiDAR odometry, inertial pre-integration (IMU), and GNSS constraints within a factor graph optimization framework for robust state estimation.

Datasets

The authors evaluated their system on two representative railway datasets collected from real-world railway lines in Hefei and Inner Mongolia. These datasets are characterized by varying durations, travel distances, and include segments with temporary GNSS outages. Ground truth trajectories were generated using post-processing kinematic (PPK) analysis combined with RTK corrections.

Key Insights and Ideas

  • Semantic Stability is Key: The paper highlights that simply using semantic labels isn't enough. Understanding the stability and reliability of different semantic categories (e.g., poles are more stable than vegetation) is crucial for robust SLAM in railway environments.
  • Adaptive Mapping Outperforms Fixed Grids: The adaptive voxel map, which adjusts its resolution based on local geometry and semantic information, proves significantly better at preserving critical railway infrastructure details compared to traditional fixed-grid approaches.
  • Probabilistic Filtering is More Robust: Leveraging segmentation confidence probabilities, rather than just binary labels, allows for more nuanced filtering of noisy data, leading to cleaner point clouds and more accurate registration.
  • GNSS-Denied Navigation is Achievable: By effectively integrating semantic constraints as long-term anchors, SERail-SLAM demonstrates impressive trajectory consistency even when GNSS signals are unavailable.

Key Results

  • Superior Accuracy: SERail-SLAM achieved a significantly lower RMSE (1.88 m) compared to state-of-the-art methods like FAST-LIO2 (6.09 m) and LIO-SAM (4.13 m) on the evaluated datasets.
  • Robustness in GNSS Outages: The system maintained trajectory consistency during temporary RTK outages, outperforming other methods that exhibited noticeable drift.
  • High-Quality Mapping: The generated 3D point cloud maps showed high continuity and structural completeness, accurately reconstructing details like rail-ballast geometry and overhead contact systems.
  • Real-time Performance: The system provides full state estimates at a 10 Hz sampling rate, synchronized with the LiDAR scan frequency, suitable for real-time applications.

Limitations and Future Work

The authors acknowledge that while the current work focuses on relatively flat railway terrain, more complex environments like long tunnels, deep valleys, and mountainous corridors might introduce additional challenges due to reduced geometric diversity and constrained motion. Future research directions include developing lightweight, structure-preserving semantic maps for efficient storage and leveraging semantic information for robust visual and LiDAR-based semantic relocalization capabilities.

Summary updated

Abstract

Reliable state estimation in railway environments presents significant challenges due to geometric degeneracy resulting from repetitive structural layouts and point cloud sparsity caused by high-speed motion. Conventional LiDAR-based SLAM systems frequently suffer from longitudinal drift and mapping artifacts when operating in such feature-scarce and dynamically complex scenarios. To address these limitations, this paper proposes SERail-SLAM, a robust semantic-enhanced multi-sensor fusion framework that tightly couples LiDAR odometry, inertial pre-integration, and GNSS constraints. Unlike traditional approaches that rely on rigid voxel grids or binary semantic masking, we introduce a Semantic-Enhanced Adaptive Voxel Map. By leveraging eigen-decomposition of local point distributions, this mapping strategy dynamically preserves fine-grained stable structures while compressing redundant planar surfaces, thereby enhancing spatial descriptiveness. Furthermore, to mitigate the impact of environmental noise and segmentation uncertainty, a confidence-aware filtering mechanism is developed. This method utilizes raw segmentation probabilities to adaptively weight input measurements, effectively distinguishing reliable landmarks from clutter. Finally, a category-weighted joint optimization scheme is implemented, where feature associations are constrained by semantic stability priors, ensuring globally consistent localization. Extensive experiments in real-world railway datasets demonstrate that the proposed system achieves superior accuracy and robustness compared to state-of-the-art geometric and semantic SLAM methods.
BibTeX
@inproceedings{song2026serail,
  author = {Weiwei Song and Shiqi Zheng and Xinye Dai and X. Wang and Y. N. Wang and Z. Y. Wang and Shujie Zhou and Wei Liu and Yidong Lou},
  title = {{SERail-SLAM: Semantic-Enhanced Railway LiDAR SLAM}},
  booktitle = {Machines},
  year = {2026},
  doi = {10.3390/machines14010072},
}
Figure