PGO-LIOM: Tightly Coupled LiDAR-Inertial Odometry and Mapping via Parallel and Gradient-Free Optimization
Hongming Shen, Qun Zong, Bailing Tian, Xuewei Zhang, Hanchen Lu
Abstract
Real-time localization and mapping for micro aerial vehicles (MAV) is a challenging problem, due to the limitation of the onboard computational power. In this article, a tightly coupled light detection and ranging (LiDAR)-inertial odometry is developed, which achieves high accuracy, real-time trajectories estimation for MAV utilizing only onboard sensors and a low-power onboard computer. The key idea of the proposed method is to integrate the IMU measurements, correct LiDAR matching measurements, LiDAR matching outliers into one nonlinear and noncontinuous objective function, and formulate the localization and mapping problem as a stochastic optimization problem. To deal with the nonlinear and noncontinuous objective function, a gradient-free optimization method is proposed to solve the stochastic optimization problem with a single parallel iteration. The novel constructed objective function and gradient-free optimization algorithm enable the proposed LiDAR-inertial odometry to achieve high accuracy and low time consumption. The effectiveness of the proposed method is demonstrated through various scenarios, including public datasets and real-world flight experiments.