OW-LOAM: Observation-Weighted LiDAR Odometry and Mapping

Abstract

Simultaneous Localization and Mapping (SLAM) is essential for robots, especially in unfamiliar indoor environments where other localization methods such as GNSS, UWB are unavailable. LOAM, as a state-of-the-art LiDAR SLAM method, works by extracting corner and surf points from raw point clouds and matching them with accumulated maps. However, the bisquare weight it uses for each observation is derived from the observation residual, which cannot reflect the actual observation quality and is of little help in improving the system accuracy. In this paper, we propose a novel method termed OW-LOAM, which takes the difference in the observation qualities into account by replacing the bisquare weight in LOAM with the inverse of the estimated variance of the observation noise based on Bayesian estimation theory. We conduct a series of experiments in various indoor environments of different scales, and the results show that the proposed OW-LOAM outperforms the original LOAM in both accuracy and robustness.