Reliable Graph-Slam Framework to Generate 2D LIDAR Intensity Maps for Autonomous Vehicles

Abstract

This paper proposes a Graph-Slam framework to increase the map position accuracy in critical environments. The road is divided into nodes to encode the road surface based on LIDAR reflectivity. This strategy allows to apply Phase Correlation to estimate the relative positions between the nodes precisely. In addition, the tactic to identify nodes in the global coordinate system enables to design the cost function with integrating sequential and anchoring edges for each node. This prevents any deviation in the road context and improves the consistency and the global position accuracy of the map especially in the revisited areas. Many particular issues such as processing time, edge calculation and covariance estimation are highlighted as well. The experimental results have verified the robustness, simplicity and reliability of the proposed framework to generate precise and largescale maps that can safely be used for localizing autonomous vehicles against expensive GNSS/INS-RTK generated maps.