IN2LAAMA: INertial Lidar Localisation Autocalibration And MApping (1905.09517v3)

Abstract: In this paper, we present INertial Lidar Localisation Autocalibration And MApping (IN2LAAMA): an offline probabilistic framework for localisation, mapping, and extrinsic calibration based on a 3D-lidar and a 6-DoF-IMU. Most of today's lidars collect geometric information about the surrounding environment by sweeping lasers across their field of view. Consequently, 3D-points in one lidar scan are acquired at different timestamps. If the sensor trajectory is not accurately known, the scans are affected by the phenomenon known as motion distortion. The proposed method leverages preintegration with a continuous representation of the inertial measurements to characterise the system's motion at any point in time. It enables precise correction of the motion distortion without relying on any explicit motion model. The system's pose, velocity, biases, and time-shift are estimated via a full batch optimisation that includes automatically generated loop-closure constraints. The autocalibration and the registration of lidar data rely on planar and edge features matched across pairs of scans. The performance of the framework is validated through simulated and real-data experiments.

• The paper presents a probabilistic framework integrating inertial preintegration with GP regression to correct motion distortion in sequential lidar scans.
• It enhances state estimation by tightly coupling feature extraction, data association, and batch optimization to improve map accuracy in dynamic environments.
• Results achieve under 0.1m positional RMSE and automatically estimate calibration parameters, eliminating the need for dedicated calibration targets.

IN2LAAMA: An Inertial Lidar Framework for Mapping and Autocalibration

The paper "IN2LAAMA: INertial Lidar Localisation Autocalibration And MApping" introduces an offline probabilistic framework designed for accurate localization, mapping, and extrinsic calibration leveraging a 3D-Lidar system combined with a 6-Degree-of-Freedom Inertial Measurement Unit (IMU). This approach focuses on addressing the inherent issue of motion distortion in lidar data due to inaccurate sensor trajectory estimation during scan acquisition. This distortion arises because lidar data points are acquired sequentially over time during a scan, not simultaneously, thereby requiring knowledge of the sensor's path for correction.

The authors propose a novel method integrating inertial preintegration with a continuous representation of observed data, enhancing trajectory estimation precision while bypassing some traditional motion model assumptions. This method assesses system motion at any instance, thanks to Upsampled Preintegrated Measurements (UPMs), which allow inertial data interpolation at arbitrary timestamps using Gaussian Process (GP) regression. This strategy anchors the precise correction of motion distortion, enhancing the mapping accuracy without explicitly modeling the system motion analytically.

The IN2LAAMA framework comprises both backend and frontend components, emphasizing a tight integration between feature extraction, data association, and batch optimization for state estimation. This integration involves recalculating features based on the latest state estimates, thereby iteratively improving map accuracy. In particular, the framework uses point-to-plane and point-to-edge constraints to map lidar data and maximize registration reliability. Sensor biases and inter-sensor time-shifts are also estimated as part of the optimization, modeled as Brownian motion and Gaussian distributions respectively.

Through rigorous simulated and real-data experiments, the IN2LAAMA system demonstrates superior performance in 3D mapping tasks, especially in dynamic and indoor environments, compared to contemporary methods like LOAM. By leveraging inertial data tightly integrated into the state estimation, the framework shows significant improvements in highly dynamic scenarios, making it robust against estimation failures that occur in feature-poor scans. Furthermore, the calibration process eliminates the need for dedicated calibration targets by automatically estimating calibration parameters alongside mapping, thus enabling seamless deployment without prior calibration environments.

Results indicate IN2LAAMA’s high accuracy and reliability in complex motion scenarios with the integration of IMU factors, validated by simulated trials and real-world evaluations performed indoors and outdoors, achieving less than 0.1 meters positional RMSE. The ability to estimate robust and targetless calibration parameters concurrently during operation further heightens usability in dynamic mapping applications.

Despite the comprehensive functionality, the authors acknowledge computational time and memory constraints due to the cubic complexity of GP regression and the extensive use of UPMs. Future work includes potential real-time operations through optimization techniques like sliding window estimations or leveraging GPU parallel computation, alongside improving loop closure mechanisms and examining alternative data representations such as surfels for unstructured environments.

Ultimately, the IN2LAAMA framework represents a rigorous and precise method for lidar-inertial integration and mapping, achieving significant strides in addressing motion distortion and calibration challenges, ultimately benefitting a multitude of autonomy-focused applications where accuracy and adaptability are critical.