GRIL-Calib: Targetless Ground Robot IMU-LiDAR Extrinsic Calibration Method using Ground Plane Motion Constraints

TL;DR

This work tackles the challenge of targetless IMU-LiDAR extrinsic calibration for ground robots whose motion is constrained to a plane, which traditionally leads to unobservable directions and degraded calibration. It introduces GRIL-Calib, a framework that combines ground-plane information with a ground plane motion constraint and a single optimization to recover full 6-DoF extrinsics, including unobservable directions. The method enhances LiDAR odometry with a ground-plane residual, synchronizes IMU and LiDAR data, and jointly optimizes rotation, translation, time offset, and biases using three residuals in one MAP-like objective, achieving superior accuracy and robustness across multiple real-world datasets. The approach is openly available and demonstrates practical impact for reliable sensor fusion in ground robotics, though it acknowledges sensitivity to uneven ground that warrants future work.

Abstract

Targetless IMU-LiDAR extrinsic calibration methods are gaining significant attention as the importance of the IMU-LiDAR fusion system increases. Notably, existing calibration methods derive calibration parameters under the assumption that the methods require full motion in all axes. When IMU and LiDAR are mounted on a ground robot the motion of which is restricted to planar motion, existing calibration methods are likely to exhibit degraded performance. To address this issue, we present GRIL-Calib: a novel targetless Ground Robot IMU-LiDAR Calibration method. Our proposed method leverages ground information to compensate for the lack of unrestricted full motion. First, we propose LiDAR Odometry (LO) using ground plane residuals to enhance calibration accuracy. Second, we propose the Ground Plane Motion (GPM) constraint and incorporate it into the optimization for calibration, enabling the determination of full 6-DoF extrinsic parameters, including theoretically unobservable direction. Finally, unlike baseline methods, we formulate the calibration not as sequential two optimizations but as a single optimization (SO) problem, solving all calibration parameters simultaneously and improving accuracy. We validate our GRIL-Calib by applying it to various real-world datasets and comparing its performance with that of existing state-of-the-art methods in terms of accuracy and robustness. Our code is available at https://github.com/Taeyoung96/GRIL-Calib.

Figures (7)

• Figure 1: Illustration of a ground robot equipped IMU ($I$) sensor and LiDAR ($L$) sensor. We leverage the IMU height from the ground ($\hat{d}_I$) as minimal prior knowledge and the LiDAR ground segmentation points ($\mathcal{G}$) for IMU-LiDAR extrinsic calibration.
• Figure 2: The overview of proposed GRIL-Calib system. During the optimization stage (Green Box), we use 3 constraints. See Section \ref{['optimization']} for more details.
• Figure 3: Visualization of timestamp for each IMU measurement and LiDAR Odometry.
• Figure 4: Illustration of the geometric rotation relationship between ground and each sensor. This robot platform is used in the M2DGR dataset.
• Figure 5: Visualization for each of the extrinsic parameters. The dotted line represents the ground truth. Closer to the dotted line represents the best estimation.