From Developer Needs to Domain-Specific Languages: Specifying Sceneries for Simulation-Based Validation of Mobile Robots

Ensuring the robustness and reliability of robotic navigation systems is critical for their deployment in dynamic real-world environments. While some developers employ simulation-based testing to validate their systems, many refrain due to the time investment required to create virtual environments using traditional CAD or 3D modeling tools. Procedural scenery generation offers automated synthesis of diverse environments, but current tools lack precise control over the generation process. To address this, we revisit two previously proposed domain-specific languages (DSLs): the FloorPlan DSL and the Variation DSL, designed to specify and vary indoor environments. The core contribution of this paper is the systematic design process of these DSLs, informed by expert developer interviews, which directly influenced language features and design choices. We also present tools that translate DSL models into artifacts compatible with the Gazebo simulator. Our evaluation, through experiments introducing structural elements like columns and doors, assesses the modeling effort and the impact on localization performance. Our results confirm that additional details improve the localization accuracy, and show the tradeoff between modelling complexity and increased performance.