Multi-stage Perception, Positioning and Planning for Automatic Wireless Charging of AGVs

The developing wireless charging technology has great potential for application on Automated Guided Vehicles (AGV). However, due to the strict requirements for high-efficiency wireless charging, it is difficult to navigate towards the charging pile at one go, especially under large-scale semi-structured or unstructured scenes where there have not enough features to obtain precise enough global positioning. Furthermore, single-stage planning relies largely on AGV's control performance, which often leads to failed alignment. In this paper, we discuss the factors that impact wireless charging efficiency and determine the required positioning accuracy. Accordingly, a multi-stage perception, positioning and planning method composed of coarse positioning, dynamic adjustment, and end-to-end alignment is proposed. In the coarse positioning stage, the use of iterative closest point (ICP) and simultaneous localization and mapping (SLAM) allows the AGV to relocate and reach the designated area flexibly and quickly. In the dynamic adjustment stage, AprilTag and LiDAR inertial odometry (LIO) is fused to estimate AGV's pose more accurately by addressing problems with AprilTag exceeding camera's field of view (FOV) or large observation errors at long distance. Finally, high-precision positioning information is obtained in the end-to-end alignment stage with coupled monocular camera and 2D LiDAR measurements. In each stage, different path constraint and obstacle avoidance strategy for motion planning are adopted to achieve higher efficiency. We show that with the proposed method, automatic wireless charging tasks could be performed without the need of readjustment, even under unstructured environment.