I2EKF-LO: A Dual-Iteration Extended Kalman Filter Based LiDAR Odometry (2407.02190v1)

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.

Insightful Overview of the Dual-Iteration Extended Kalman Filter Based LiDAR Odometry (I<sup\>2</sup>EKF-LO)

The paper under review introduces a sophisticated approach to LiDAR odometry, focusing on the challenges posed by motion distortion in LiDAR point clouds and the determination of process noise within the traditional Iterative Extended Kalman Filter (IEKF) framework. The authors present a Dual-Iteration Extended Kalman Filter (I<sup\>2</sup>EKF) method that advances LiDAR odometry through both methodological innovation and empirical validation.

Core Contributions

The paper identifies critical shortcomings of standard IEKF-based LiDAR odometry, namely the partial iteration over observation equations and static process noise assumptions. In response, the authors propose the I<sup\>2</sup>EKF-LO framework, which integrates novel enhancements to address motion distortions and adapt process noise dynamically:

1. Dual Iteration Process: The I<sup\>2</sup>EKF framework introduces dual iterations, focusing on the correction of motion distortion and observation processes. This dual approach ensures refined point cloud quality and state estimation precision by iteratively manipulating both the observation and prediction processes.
2. Dynamic Process Noise Adjustment: Leveraging measurement innovation, the framework dynamically adjusts process noise, enhancing system robustness across varying motion intensities. The adjustment mechanism is a significant departure from the constant noise assumption typically embedded within traditional frameworks.
3. Incorporation of SE(3) Transformations: By accommodating rotational and translational coupling, I<sup\>2</sup>EKF facilitates accurate management of transformations within the LiDAR coordinate system, adapting seamlessly to diverse sensor carrier motion models.
4. Extensive Empirical Validation: The paper provides comprehensive empirical results demonstrating that I<sup\>2</sup>EKF-LO achieves superior accuracy and computational efficiency compared to existing algorithms such as KISS-ICP, CT-ICP, and F-LOAM across a variety of datasets.

Technical Insights and Experimental Results

The I<sup\>2</sup>EKF framework's iteration over prediction processes distinguishes it from conventional methods by effectively minimizing input point cloud distortion. Empirical validation shows marked improvements, particularly in scenarios where traditional IEKF-LO would exhibit significant drift due to motion artifacts. The paper reports substantial accuracy advancements: I<sup\>2</sup>EKF-LO consistently outperforms traditional IEKF implementations, especially in environments characterized by erratic or high-intensity movements.

The comparative analysis detailed in the experiments section underscores I<sup\>2</sup>EKF-LO's computational efficiency. Despite the added complexity of dual iterations, the framework maintains competitive processing speeds, attributed to its leveraging of direct methods and ikd-tree structures for map management.

Implications and Future Developments

Practically, the dual-iteration approach enhances the reliability of LiDAR odometry systems within autonomous vehicles and robotics, particularly in dynamic environments where traditional assumptions of constant motion are invalidated. Theoretically, the paper paves the way for further integration of adaptive noise models in state estimation processes, offering a robust template for future research in dynamic system modeling.

The potential extension of I<sup\>2</sup>EKF to LiDAR-Inertial Odometry (LIO) systems could harness IMU bias updates, enhancing the distortion correction scope. However, given the marginal improvement over traditional implementations observed empirically, further research is necessary to optimize computational overhead versus accuracy trade-offs.

In conclusion, this paper presents a valuable advancement in the field of LiDAR odometry, substantiated by rigorous empirical testing and theoretical clarity. It effectively addresses persistent challenges in motion distortion correction and adaptive process noise modeling, with promising implications for both practical applications and theoretical advancements in nonlinear filter-based odometry.