Robust LiDAR SLAM in Geometrically Degenerate Environments

You are reading version 1 of this direction, frozen 2026-10-04. The latest version is version 2.

Version 1 of 2195 papers267 conceptsfrozen 2026-10-04

Simultaneous Localization and Mapping (SLAM) is often described as a solved problem in robotics, yet it frequently fails in the most mundane settings: a long, smooth corridor, a perfectly cylindrical tunnel, or a wide-open field. These failures stem from a phenomenon known as geometric degeneracy, where the environment lacks sufficient structural variation to uniquely determine a robot's position.

The Geometry of Failure

To understand why SLAM fails in a corridor, we must look at how LiDAR registration works. Most modern systems, such as those discussed by Wei Xu in FAST-LIO2, rely on matching current laser points to a previously built map or a set of geometric primitives like planes and edges.

Mathematically, this is an optimization problem. The system tries to find a transformation (rotation and translation) that minimizes the distance between points and their corresponding surfaces. In a feature-rich room with corners, desks, and windows, the surface normals point in many different directions, providing "constraints" for all six degrees of freedom (DoF).

However, as Li Li and Jian Yao observe, in a long uniform corridor, almost all surface normals are perpendicular to the direction of travel. There is no geometric "anchor" to stop the algorithm from "sliding" along the axis of the hallway. This leads to massive longitudinal drift, where the robot thinks it has moved five meters when it has actually moved ten. Similar issues occur in tunnels, where Jiun Fatt Chow notes that cylindrical symmetry makes it impossible for scan-matching algorithms to distinguish between different positions along the tunnel's axis.

Detecting and Quantifying Localizability

The first step in building a robust system is knowing when to trust the sensor. Turcan Tuna and Marco Hutter introduced X-ICP, which analyzes the "geometric strength" of sensor data in real-time. By examining the eigenvalues of the optimization matrix, the system can identify which directions are well-constrained and which are "degenerate."

When degeneracy is detected, the system can switch strategies. Instead of forcing the LiDAR to estimate movement in an unconstrained direction—which would just result in the system chasing sensor noise—the filter can "ignore" the LiDAR for that specific axis and rely on other sensors. Jie Xu and colleagues formalize this in their Selective Kalman Filter, which selectively integrates auxiliary data only when the primary geometric constraints fail.

Beyond Geometry: Intensity and Multi-Sensor Fusion

If the shape of the environment is not enough, researchers look to other properties of the signal. Chen Wang and Han Wang developed Intensity-SLAM, which uses the reflectivity (intensity) of the laser return. Even if a wall is perfectly flat, it might have posters, signs, or different paint types that create "visual" landmarks in the intensity channel. Patrick Pfreundschuh expanded on this with COIN-LIO, treating intensity as a virtual camera to provide constraints where geometry is smooth.

Other researchers focus on hardware and environmental constraints:

  • Ground Constraints: Wei Xin and Jixin Lv developed Ground-SLAM, which treats the floor as a persistent geometric constraint to eliminate vertical drift in multi-floor buildings.
  • Multi-LiDAR Systems: Jianhao Jiao shows in M-LOAM that using multiple LiDARs with different orientations can "see" around degeneracy; while one sensor faces a smooth wall, another might see a distant doorway.
  • External Anchors: In agricultural settings where rows of crops look identical, Mateus Valverde Gasparino uses CropNav to switch between LiDAR and GNSS (GPS) depending on the localizability of the environment.

Continuous-Time Trajectories

A final frontier in robustness involves how we model the robot's path. Traditional SLAM treats a "sweep" of LiDAR points as a single frozen moment in time. However, Jianke Zhu and Xin Zheng argue in Traj-LO that parameterizing movement as a continuous-time trajectory provides inherent kinematic smoothness. This mathematical "stiffness" helps the system resist the urge to jump or slide when geometric features become sparse, maintaining a plausible path even through featureless voids.

By combining these approaches—detecting mathematical degeneracy, leveraging intensity data, and enforcing physical motion constraints—researchers are moving SLAM from "lab-ready" to "world-ready," capable of navigating the featureless tunnels and corridors that once acted as robotic dead zones.

Papers in this version

Showing 60 of 195 papers.

History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 230 papers
  2. Version 1 · 2026-10-04 · 195 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_geometric_degeneracy_geometric_feature_d_346ba907,
  title = {geometric degeneracy (geometric feature dependency)},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/geometric-degeneracy-geometric-feature-dependency/rdr_346ba90725ee4876a680c46e2b02f50a},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_robust_lidar_slam_in_geometrically_degen_346ba907_v1,
  title = {Robust LiDAR SLAM in Geometrically Degenerate Environments},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/robust-lidar-slam-in-geometrically-degenerate-environments/rdr_346ba90725ee4876a680c46e2b02f50a/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}