Self-supervised Geometry Reasoning for LiDAR Simultaneous Localization and Mapping

2026-06-29

AI Summary

Traditional LiDAR SLAM systems, which rely on handcrafted geometric estimators, are highly susceptible to noise in scenarios with sparse or low-resolution point clouds. This work proposes a self-supervised framework that models each point as a Gaussian distribution to explicitly represent local geometry and introduces a signed geometric consistency constraint to enable learning without ground-truth labels. To the best of our knowledge, this is the first approach to achieve self-supervised local geometric estimation without requiring annotated data. The method integrates seamlessly with existing SLAM backends and establishes a recursive optimization loop between geometric reasoning and pose estimation, allowing mutual refinement. Evaluated on the KITTI dataset across multiple LiDAR resolutions, the proposed approach significantly improves odometry accuracy and global registration performance.

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.

Problem

• LiDAR SLAM
• local geometry
• self-supervised learning
• covariance estimation
• geometry reasoning

Innovation

• self-supervised learning
• local geometry representation
• LiDAR SLAM
• Gaussian point representation
• recursive refinement