[Paper Review] LIO-SAM: Tightly-coupled Lidar Inertial Odometry via Smoothing and Mapping

Tixiao Shan, Brendan Englot+4 more|arXiv (Cornell University)|Jul 1, 2020

Robotics and Sensor-Based Localization24 references99 citations

TL;DR

LIO-SAM builds a tightly-coupled lidar–IMU framework on a factor graph to provide real-time trajectory estimation and mapping, incorporating loop closures and optional GPS for drift correction.

ABSTRACT

We propose a framework for tightly-coupled lidar inertial odometry via smoothing and mapping, LIO-SAM, that achieves highly accurate, real-time mobile robot trajectory estimation and map-building. LIO-SAM formulates lidar-inertial odometry atop a factor graph, allowing a multitude of relative and absolute measurements, including loop closures, to be incorporated from different sources as factors into the system. The estimated motion from inertial measurement unit (IMU) pre-integration de-skews point clouds and produces an initial guess for lidar odometry optimization. The obtained lidar odometry solution is used to estimate the bias of the IMU. To ensure high performance in real-time, we marginalize old lidar scans for pose optimization, rather than matching lidar scans to a global map. Scan-matching at a local scale instead of a global scale significantly improves the real-time performance of the system, as does the selective introduction of keyframes, and an efficient sliding window approach that registers a new keyframe to a fixed-size set of prior ``sub-keyframes.'' The proposed method is extensively evaluated on datasets gathered from three platforms over various scales and environments.

Motivation & Objective

• Address drift and accuracy limitations in lidar odometry for real-time mobile robot SLAM.
• Enable tight fusion of LiDAR, IMU, and absolute sensors (e.g., GPS) within a single factor-graph framework.
• Achieve real-time performance through local scan-matching, selective keyframes, and a sliding-window approach.
• Support loop closures and place recognition to correct long-term drift and enable mapping across diverse environments.
• Validate the approach across multiple platforms and datasets to demonstrate robustness and scalability.

Proposed method

• Formulate lidar-inertial odometry on a factor graph with IMU preintegration, lidar odometry, GPS, and loop closure factors.
• Use IMU preintegration to de-skew lidar scans and provide an initial motion guess for lidar odometry optimization, while jointly estimating IMU biases.
• Adopt a local, map-based scan-matching approach using a fixed-size set of sub-keyframes within a sliding window for real-time performance.
• Construct a two-type voxel map from the most recent keyframes to support edge and planar feature matching.
• Incorporate GPS and loop-closure factors when available to reduce drift and improve global consistency.
• Utilize a Bayes-tree based incremental optimization (iSAM2) on a global trajectory-factor graph and marginalize old lidar scans to maintain real-time performance.

Experimental results

Research questions

• RQ1 Can a tightly-coupled lidar–IMU system built on a factor graph deliver real-time, low-drift trajectory estimation and mapping across diverse environments?
• RQ2 How do local (not global) scan-matching, keyframe strategies, and marginalization affect real-time performance and accuracy compared to dense, global approaches?
• RQ3 What is the impact of adding absolute measurements (GPS) and loop closures on drift correction and map consistency?
• RQ4 How does LIO-SAM compare to LOAM and LIOM in terms of accuracy and runtime under challenging motions?

Key findings

• LIO-SAM achieves real-time performance with significantly lower per-scan runtimes than LOAM and LIOM across multiple datasets.
• The method yields maps that are consistent with external references (e.g., Google Earth) even without GPS in some cases.
• LIO-SAM remains robust under aggressive rotations and translations, outperforming LOAM and LIOM in several scenarios.
• GPS and loop-closure factors effectively reduce drift; LIO-SAM can close loops where GPS alone cannot, improving global consistency.
• The system supports substantial throughput (stress-tested up to 13x real-time data) while maintaining accuracy and real-time operation.