CLINS: Continuous-Time Trajectory Estimation for LiDAR-Inertial System (2109.04687v1)

Abstract: In this paper, we propose a highly accurate continuous-time trajectory estimation framework dedicated to SLAM (Simultaneous Localization and Mapping) applications, which enables fuse high-frequency and asynchronous sensor data effectively. We apply the proposed framework in a 3D LiDAR-inertial system for evaluations. The proposed method adopts a non-rigid registration method for continuous-time trajectory estimation and simultaneously removing the motion distortion in LiDAR scans. Additionally, we propose a two-state continuous-time trajectory correction method to efficiently and efficiently tackle the computationally-intractable global optimization problem when loop closure happens. We examine the accuracy of the proposed approach on several publicly available datasets and the data we collected. The experimental results indicate that the proposed method outperforms the discrete-time methods regarding accuracy especially when aggressive motion occurs. Furthermore, we open source our code at \url{https://github.com/APRIL-ZJU/clins} to benefit research community.

• The paper introduces a continuous-time framework that fuses asynchronous LiDAR and IMU data using B-splines to mitigate motion distortion.
• It employs a non-rigid registration and a two-stage correction mechanism to enhance loop closure detection and reduce RMSE in translation and rotation.
• Experimental results demonstrate significant accuracy improvements over discrete-time methods, enabling robust 3D reconstruction even with 2D LiDAR systems.

Continuous-Time Trajectory Estimation for LiDAR-Inertial Systems

The research paper entitled "CLINS: Continuous-Time Trajectory Estimation for LiDAR-Inertial System" presents an approach for enhancing Simultaneous Localization and Mapping (SLAM) accuracy by developing a continuous-time trajectory estimation framework utilizing LiDAR and Inertial Measurement Unit (IMU) data. This paper is pertinent to disciplines that leverage SLAM, including autonomous navigation, scene reconstruction, and mixed reality, given its improvements over existing discrete-time methodologies for trajectory estimation.

Framework and Methodology

The core contribution of the paper is the introduction of a continuous-time trajectory estimator tailored for the fusion of high-frequency and asynchronous sensor data. The proposed system uniquely employs a non-rigid registration method that mitigates motion distortion in LiDAR scans. It also introduces a two-stage correction mechanism to efficiently address the global optimization challenges that arise during loop closure detection. This process ensures both accuracy and computational efficiency—a challenging balance in real-world SLAM applications.

Additionally, continuous-time trajectory estimation is implemented using B-splines to parameterize translations and rotations separately. B-splines provide smoothness and allow for the integration of asynchronous sensor data, which is beneficial when fusing LiDAR and IMU measurements. This continuous representation overcomes the limitations seen in discrete-time methods where interpolation errors are prevalent due to inconsistent sensor frequencies. Continuous trajectory representation ensures high temporal resolution, crucial for realizing precision in scenarios involving aggressive motion.

Experimental Evaluation

The efficacy of the proposed framework is demonstrated through comprehensive evaluations on publicly available datasets as well as custom-collected datasets. Experimental results highlight that CLINS outperforms comparable discrete-time methods, such as LOAM, LIOM, and LIO-SAM, particularly during high-dynamic motions where traditional methods typically show higher errors. Notably, CLINS achieved reduced Root Mean Square Errors (RMSE) in both translation and rotation estimations across various scenarios, as evidenced in the result tables provided in the paper.

A notable aspect of the presented system is its applicability beyond traditional 3D systems, extending functionality to 2D LiDAR systems which typically face limitations in estimating six-dimensional poses. By leveraging the continuous-time trajectory, the authors demonstrated high accuracy in constructing three-dimensional reconstructions with a 2D LiDAR system, showcasing potential novel applications enabled by the global trajectory estimation.

Implications and Future Directions

The findings of this paper have significant implications for the field of robotic perception and navigation. By providing seamless integration of asynchronous data sources, this approach sets a benchmark for high-precision, real-time trajectory estimation in dynamic environments. The open sourcing of the CLINS framework also holds promise for advancing collaborative research efforts and facilitating further innovations in SLAM methodologies.

Future avenues of exploration could delve into refining the computational efficiency and accuracy of the system. Potential directions include optimizing the number of static control points in the B-spline representation or deriving analytical Jacobians in place of automatic derivations to expedite iterative computations. Additionally, broader implementation across different SLAM platforms and further tests in diverse environmental conditions would offer deeper insights into the robustness and scalability of the CLINS system.

In conclusion, this paper presents a pivotal step in the evolution of SLAM technology, providing enriched frameworks and methodologies that will likely inspire subsequent research and development endeavors.