A Comprehensive Evaluation of LiDAR Odometry Techniques

2025-07-21

AI Summary

A systematic, component-level comparison of LiDAR odometry (LO) pipelines is currently lacking, hindering the design of high-accuracy and robust systems. This work presents the first large-scale ablation study across the full LO stack—spanning multiple datasets (KITTI, nuScenes), LiDAR types (mechanical and solid-state), environments (urban, highway, dynamic scenes), and motion patterns. We unify classical and state-of-the-art algorithms for point cloud registration, feature extraction, and optimization-based estimation into a reproducible, benchmarkable LO analysis framework. Through empirical evaluation, we quantify each module’s contribution to localization accuracy and robustness, identify optimal component combinations, and validate their superior performance under diverse real-world conditions. This study fills a critical gap in system-level LO evaluation and establishes the first empirically grounded benchmark for algorithm selection and end-to-end LO system design.

Technology Category

• Intelligent Robots: Localization, Mapping, and Navigation
• Search and Optimization: Algorithm Configuration
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Security and Privacy: Large-scale security measurements

Abstract

Light Detection and Ranging (LiDAR) sensors have become the sensor of choice for many robotic state estimation tasks. Because of this, in recent years there has been significant work done to fine the most accurate method to perform state estimation using these sensors. In each of these prior works, an explosion of possible technique combinations has occurred, with each work comparing LiDAR Odometry (LO) "pipelines" to prior "pipelines". Unfortunately, little work up to this point has performed the significant amount of ablation studies comparing the various building-blocks of a LO pipeline. In this work, we summarize the various techniques that go into defining a LO pipeline and empirically evaluate these LO components on an expansive number of datasets across environments, LiDAR types, and vehicle motions. Finally, we make empirically-backed recommendations for the design of future LO pipelines to provide the most accurate and reliable performance.

Problem

• Evaluating LiDAR odometry techniques for state estimation
• Comparing building-blocks of LiDAR odometry pipelines
• Recommending designs for accurate LiDAR odometry performance

Innovation

• Evaluates LiDAR odometry components across diverse datasets
• Compares various LO pipeline building-blocks empirically
• Provides data-driven recommendations for LO pipeline design