Vision-aided Localization and Navigation for Autonomous Vehicles

Abstract

Automated Guided Vehicles (AGVs) have played an import part in many areas, such as transporting and delivery, domestic cleaning, search and rescue. In many industrial scenarios, 2D LiDAR is a popular localization and navigation device due to its relatively low cost and high accuracy. However, AGVs equipped with 2D LiDAR are unable to detect obstacles in 3D space which limits their applications in complex environment. Vision-based approaches are able to solve this problem, however, they are not efficient in map building and navigation capability. To enhance the intelligence and capabilities of traditional AGVs equipped with 2D LiDAR sensors and make it more robust in various environments, we propose a vision-aided localization and navigation system which integrates the advantages of camera and 2D LiDAR. We propose an efficient obstacle detection method in 3D space and reflect them in a 2D map produced by a 2D LiDAR. A model predictive control (MPC) based path planning is then utilized for autonomous navigation with collision free capability. Experiments with industrial AGVs in warehouse demonstrate that the AGVs are able to localize accurately and navigate with collision avoidance capability with our proposed method.