A large part of Kollmorgen’s business focuses on automating warehouses. This study focused on going beyond autonomous navigation by providing a means to automatically label parts of a warehouse by using machine-learning techniques.
Common map representations used for autonomous robot navigation only contains geometrical structures, such as a grid that encodes free and occupied areas. In this work the focus is to learn and classify commonly occurring structures (e.g. pallets, shelves), but also to detect salient regions that could be useful for estimating the initial global positions of transport robots.