PFI:AIR - TT: AGV-3D: A Low-Cost, Infrastructure-free Localization Solution for FlexibleWarehouse Automation

Lehigh University, Bethlehem PA

Investigators

Abstract

This PFI: AIR Technology Translation project focuses on translating technology to fill the need for an infrastructure free localization system to accurately estimate the position and orientation of an automated guided vehicle (AGV) in large-scale warehouse environments. The system is called AGV-3D, and it is important because it enhances current AGV technology, which in turn improves the efficiency of U.S. manufacturing by providing a reliable and safe means of meeting material movement demands. The project will result in an AGV-3D prototype being integrated into an actual AGV system, and demonstrated in a representative warehouse environment. AGV-3D has the following unique features: it integrates the latest in 3D imaging technologies to enable robust and accurate localization of an AGV vehicle in a large-scale warehouse facility without the integration of wires, magnets, or reflectors in the environment. These features provide the following advantages: lower per-vehicle cost, lower facility installation cost, and increased facility flexibility when compared to the leading competing 2D laser guidance technology in this market space. This project addresses the following technology gap(s) as it translates from research discovery toward commercial application. The first significant challenge is that the required positioning accuracy of AGVs is on the order of 1 cm. The second is that the localization system must be extremely robust as AGV warehouse installations require very high levels of availability. Unfortunately, correctly associating landmark features will be far more difficult as unlike current 2D laser guidance technologies, artificial retroreflector targets will not be installed in the environment. Thus, there are significant challenges in both data association and reconstruction. To meet these challenges, AGV-3D leverages 3D data for both mapping the environment and landmark feature segmentation. First, a 3D reconstruction of the facility is created to extract salient natural features as landmarks. Next, a map-based localization approach leverages 3D LIDAR to enable 3D feature-to-landmark matching which minimizes the potential for data association errors. By employing the latest in 3D sensor systems, AGV vehicles will be able to track these 3D features in real-time. In conjunction with the map-based localization approach, centimeter level accuracy is expected. In addition, graduate students involved in this project will receive innovation and technology translation experiences through system development, and through direct interactions with companies in the robotics, AGV, and 3D sensing spaces.