DualQuat-LOAM: LiDAR Odometry and Mapping parameterized on Dual Quaternions

Abstract

Mobile robotics is increasingly in need of low bias and computationally efficient odometry methods. In response to this need, we present a LiDAR odometry estimation approach by fully parameterizing the system using dual quaternions. To achieve this, both the features extracted from the point cloud, such as edges, surfaces and STD (Stable Triangle Descriptor) descriptors, as well as the optimizer, are represented in the dual quaternion set. This advantage allows us to incorporate both position and attitude of STD descriptors into the estimation problem, significantly improving pose estimation, which is reflected in comparative experiments with other state-of-the-art methods. Unlike traditional approaches, our odometry estimation does not rely on global maps or loop closure algorithms, further reducing computational costs. Experimental results show a translation and rotation error of 0.79% and 0.0039°/m on the KITTI dataset, with an average run time of 53 ms.