Unified Map Handling for Robotic Systems: Enhancing Interoperability and Efficiency Across Diverse Environments

TL;DR

The paper tackles the challenge of heterogeneous map formats hindering rapid deployment of robotic fleets. It proposes an open, standardized environment representation framework along with a ROS2-compatible environment_template and a converter stack to enable automatic information extraction, conversion, and procedural generation. Key contributions include a taxonomy of environmental information types, a catalog of map-encoding formats with direct/indirect data mappings, and public datasets across agricultural, urban, and indoor environments. The approach promises improved interoperability and deployment efficiency, enabling pipelines for cascaded deployments and easier collaboration within the robotics community.

Abstract

Mapping is a time-consuming process for deploying robotic systems to new environments. The handling of maps is also risk-adverse when not managed effectively. We propose here, a standardised approach to handling such maps in a manner which focuses on the information contained wherein such as global location, object positions, topology, and occupancy. As part of this approach, associated management scripts are able to assist with generation of maps both through direct and indirect information restructuring, and with template and procedural generation of missing data. These approaches are able to, when combined, improve the handling of maps to enable more efficient deployments and higher interoperability between platforms. Alongside this, a collection of sample datasets of fully-mapped environments are included covering areas such as agriculture, urban roadways, and indoor environments.

Figures (5)

• Figure 1: Collection of maps rendered using Google Earth using KML.
• Figure 2: Chord Diagram detailing examples of conversions between different standard map types. Each line indicates a conversion from the black end of the line to the destination file. Each conversion process is indicated by a different destination colour with multiple lines being used together.
• Figure 3: Regions included with the agricultural environment dataset, rendered with Google Earth. Colours of polygons indicate the type of space, with yellow showing horticultural fields, green indicating pastoral field, dark green showing forest, orange showing parkland, blue showing office buildings, cyan showing animal housing, and purple showing grower plots for specialist crops. White regions indicating connecting paths and roads across the farm.
• Figure 4: Locations across the globe in the Urban Roadway Environment dataset.
• Figure 5: Examples warehouse environments generated procedurally using wave-function collapse. Both examples had been supplied with a 5m sample template detailing an example of an how an arrangement of warehouse shelves and free space may appear on a 2D grid. (a) shows the result of a 10m generated regions, and (b) shows the result of a 15m generated region.