Tightly Coupled LiDAR-Inertial Odometry

Abstract

In the age of self-driving, LiDAR and IMU represent two of the most ubiquitous sensors in use.
Kalman filtering and loosely coupled approaches dominate industry techniques, while current research trends towards a more tighly coupled formulation involving a joint optimization of IMU and LiDAR measurements.
After two years of experience working with and creating tightly coupled LiDAR-inertial odometry (LIO) systems for offroad and indoor environments, we detail our findings regarding such implementations.
Moreover, we present a general framework involving point-to-point based registration, an adaptive robust kernel, and state-of-the-art preintegration
for odometry.
Our method operates in real-time on a moderately powerful CPU, and we showcase its capabilities in high speed offroad environments, as well as indoor environments.

Furthermore, we provide an extensive section devoted to background information required to implement our version of LiDAR-inertial odometry.
All algorithms and equations (except for IMU preintegration) are given with the goal of this document being completely self-contained.
We provide robust analysis of tradeoffs made along the way, and provide direction for future improvements based on real-world observations on real hardware.

The key contributions were developed over two programs: DARPA RACER, a program exploring high-speed off-road autonomy across desert and wooded environments,
and MMPUG, a program devoted to indoor navigation with small mobile robots.