Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

I2EKF-LO: A Dual-Iteration Extended Kalman Filter Based LiDAR Odometry

Wenlu Yu, Jie Xu, Chengwei Zhao, Lijun Zhao, Thien-Minh Nguyen, Shenghai Yuan, Mingming Bai, Lihua Xie

TL;DR

I$^2$EKF-LO introduces a Dual-Iteration Extended Kalman Filter for LiDAR odometry that undistorts the current frame while refining state estimates via iterative observation updates. It dynamically adapts process noise based on the innovation sequence and uses SE($3$) motion models to handle different sensor carriers, improving robustness across motion regimes. The approach delivers higher accuracy and efficiency than IEKF and several baselines on diverse datasets, and the code is open-source. This work advances LiDAR-only odometry by robustly compensating motion distortion and adapting to varying motion patterns without relying on IMU data.

Abstract

LiDAR odometry is a pivotal technology in the fields of autonomous driving and autonomous mobile robotics. However, most of the current works focus on nonlinear optimization methods, and still existing many challenges in using the traditional Iterative Extended Kalman Filter (IEKF) framework to tackle the problem: IEKF only iterates over the observation equation, relying on a rough estimate of the initial state, which is insufficient to fully eliminate motion distortion in the input point cloud; the system process noise is difficult to be determined during state estimation of the complex motions; and the varying motion models across different sensor carriers. To address these issues, we propose the Dual-Iteration Extended Kalman Filter (I2EKF) and the LiDAR odometry based on I2EKF (I2EKF-LO). This approach not only iterates over the observation equation but also leverages state updates to iteratively mitigate motion distortion in LiDAR point clouds. Moreover, it dynamically adjusts process noise based on the confidence level of prior predictions during state estimation and establishes motion models for different sensor carriers to achieve accurate and efficient state estimation. Comprehensive experiments demonstrate that I2EKF-LO achieves outstanding levels of accuracy and computational efficiency in the realm of LiDAR odometry. Additionally, to foster community development, our code is open-sourced.https://github.com/YWL0720/I2EKF-LO.

I2EKF-LO: A Dual-Iteration Extended Kalman Filter Based LiDAR Odometry

TL;DR

IEKF-LO introduces a Dual-Iteration Extended Kalman Filter for LiDAR odometry that undistorts the current frame while refining state estimates via iterative observation updates. It dynamically adapts process noise based on the innovation sequence and uses SE() motion models to handle different sensor carriers, improving robustness across motion regimes. The approach delivers higher accuracy and efficiency than IEKF and several baselines on diverse datasets, and the code is open-source. This work advances LiDAR-only odometry by robustly compensating motion distortion and adapting to varying motion patterns without relying on IMU data.

Abstract

LiDAR odometry is a pivotal technology in the fields of autonomous driving and autonomous mobile robotics. However, most of the current works focus on nonlinear optimization methods, and still existing many challenges in using the traditional Iterative Extended Kalman Filter (IEKF) framework to tackle the problem: IEKF only iterates over the observation equation, relying on a rough estimate of the initial state, which is insufficient to fully eliminate motion distortion in the input point cloud; the system process noise is difficult to be determined during state estimation of the complex motions; and the varying motion models across different sensor carriers. To address these issues, we propose the Dual-Iteration Extended Kalman Filter (I2EKF) and the LiDAR odometry based on I2EKF (I2EKF-LO). This approach not only iterates over the observation equation but also leverages state updates to iteratively mitigate motion distortion in LiDAR point clouds. Moreover, it dynamically adjusts process noise based on the confidence level of prior predictions during state estimation and establishes motion models for different sensor carriers to achieve accurate and efficient state estimation. Comprehensive experiments demonstrate that I2EKF-LO achieves outstanding levels of accuracy and computational efficiency in the realm of LiDAR odometry. Additionally, to foster community development, our code is open-sourced.https://github.com/YWL0720/I2EKF-LO.
Paper Structure (26 sections, 15 equations, 6 figures, 3 tables, 1 algorithm)

This paper contains 26 sections, 15 equations, 6 figures, 3 tables, 1 algorithm.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORKS
  3. OVERVIEW AND PREPROCESSING
  4. System Overview
  5. Kinematic Model
  6. Uniform Motion Model
  7. Observation Model
  8. Point Cloud Preprocessing
  9. DUAL-ITERATION EXTENDED KALMAN FILTER
  10. IEKF Review
  11. Forward Propagation in I$^2$EKF Uniform Motion Model
  12. Iterative Update in I$^2$EKF
  13. Iterative Distortion Correction Update
  14. Iterative Observation Update
  15. Process Noise Update
  16. ...and 11 more sections

Figures (6)

  • Figure 1: Mapping results of I$^2$EKF-LO in HIT-TIB dataset (sequence walk). When I$^2$EKF-LO iterates only over the observation process, it degenerates to normal IEKF-LO. I$^2$EKF-LO has better handling of details compared to IEKF-LO. While KISS-ICP fails completely on this sequence using the same resolution.
  • Figure 2: Framework of I$^{2}$EKF-LO.
  • Figure 3: The experimental platform uses the Agilex Scout Mini as the mobile chassis, featuring four-wheel differential steering. It is equipped with a Livox Mid-360 LiDAR and uses the Nvidia AGX Xavier as the computing platform.
  • Figure 4: Mapping results of I$^2$EKF-LO in NTU VIRAL dataset.
  • Figure 5: Mapping results in Urbanloco dataset. Significant $z$-axis errors in (b) CT-ICP, (c) KISS-ICP, and (d) F-LOAM as the lidar returns to the vicinity of the origin. While (a) I$^2$EKF-LO has a much smaller error in the $z$-axis direction.
  • ...and 1 more figures