Terrestrial laser scanning as a support to design and deployment of automated warehouses

The design of modern automated warehouses is a complex task that requires detailed and accurate models of industrial buildings. However, the traditional design workflow is primarily based on 2D floor plans that may contain outdated or even missing parts. The common approach to update existing 2D floor plans is to perform surveys based on sparse measures, taken by hand-held instruments, such as Laser Distance Meters, or Total Stations. Indeed, these instruments provide very accurate point-to-point measures, but the amount of data that can be acquired may not be sufficient. As a result, multiple surveys need to be performed as new requirements emerge during system development. Moreover, many unpredictable issues are usually discovered at the time of deployment, thus resulting in high costs and delays. Therefore, companies that operate in the warehouse automation business are starting to apply 3D Terrestrial Laser Scanning technology to overcome the limitations of traditional surveys. This brings new research challenges, never addressed in industrial environments before. A first contribution of this dissertation is the proposal of a novel workflow for the design of automated warehouses that improves the traditional development process by performing a 3D survey that combines a Terrestrial Laser Scanner and a Total Station. Automated warehouses include Autonomous Guided Vehicles (AGVs) that move along predefined paths, as well as fixed robot workcells. The workflow covers every phases from data collection to data processing, and exploitation. The proposed workflow can handle billions of points, acquired from thousands of scan stations, and it consists of several automatic and semi-automatic steps. The first step of the workflow is to perform a 3D survey by following a procedure that achieves a good trade-off between survey time, accuracy, and level of detail of the acquired point cloud. Then, high level information about the environment are extracted, exploiting innovative algorithms for large scale point cloud processing. In particular, novel approaches for ground segmentation, floor plan generation and real-time AGV collision detection are presented. The proposed approach for ground segmentation does not assume the presence of a dominant plane and it scales linearly with the number of scans. The algorithm for floor plan generation is based on the segmented ground, so that no assumptions are made about the maximum expected slope. Moreover, walls are neither required to be planar nor having orthogonal intersections, like in previous works. AGV paths can, therefore, be defined based on the generated floor plan and a real-time collision detection algorithm is proposed to verify their feasibility. Virtual Reality is also supported to provide immersive visualization. Finally, a novel approach for scan position optimization is investigated that exploits a realistic sensor model that simulates a number of fixed parameters having a strong influence on the laser measurements, like laser height from the ground, resolution, sensor range and angle of incidence of beams on both walls and ground. While some parts of the proposed workflow have been developed to solve specific problems of the warehouse automation industry, most of the developed algorithms, such as automatic ground segmentation, floor plan generation and scan position optimization can be applied in any indoor environment. All the solutions developed in this thesis have been fully integrated with existing softwares to speed up the deployment phase. Experiments have shown that the proposed workflow drastically speeds up development and deployment of system installations.