Self-supervised Geometry Reasoning for LiDAR Simultaneous Localization and Mapping
2026-06-29 • Robotics
Robotics
AI summaryⓘ
The authors present a new way to improve LiDAR SLAM, which helps robots figure out their position and map their surroundings. They teach the system to better understand local shapes by representing each point as a Gaussian distribution, capturing uncertainties naturally. Their method learns without needing detailed labels by using patterns in the data itself, and then uses this learned information to make SLAM more accurate in a feedback loop. This approach works with existing SLAM systems and showed improvements on real-world datasets with different LiDAR qualities.
LiDARSLAMGaussian distributionCovarianceSelf-supervised learningLocal geometryPoint cloudOdometryGlobal registrationKITTI dataset
Authors
Jiwoo Kim, Jinwoo Lee, Woojae Shin, Giseop Kim, Hyondong Oh
Abstract
LiDAR simultaneous localization and mapping (SLAM) relies on local geometric quantities such as covariances, correspondences, and surface structures. However, most existing pipelines rely on hand-crafted estimates of local geometry and use them as fixed inputs to LiDAR SLAM, which can make the estimated local geometry noisy and unstable in sparse regions of a point cloud or when using low-resolution LiDAR. To address this issue, this paper introduces a self-supervised framework that learns an explicit symbolic representation of local geometry and uses it to improve LiDAR SLAM recursively. Specifically, each point is represented as a Gaussian distribution, allowing local geometry to be described by a covariance. Without dense geometry labels or ground-truth poses, the framework learns by maximizing the likelihood of local geometry, with self-supervision derived from consistency relations over symbolic geometric representations, including predicted covariances, correspondences, and trajectory from SLAM. The learned geometry is then fed back into LiDAR SLAM, forming a reciprocal loop in which improved geometry enhances localization and mapping, and improved localization provides cleaner supervision for subsequent geometry reasoning. This framework is backend-agnostic and can be plugged into existing LiDAR SLAM pipelines without architectural changes. Experiments on KITTI under varying LiDAR resolutions show that the proposed method improves both odometry and global registration.