BIMgent Framework for Autonomous Building Modeling

This technical report introduces the BIMgent framework, designed for autonomous building modeling within the Architecture, Engineering, and Construction (AEC) sector. It addresses the limitations of existing computer-use agents that primarily focus on general-purpose automation, highlighting the need for specialized solutions in BIM authoring software. The report details the framework's hierarchical planning method, which decomposes the modeling workflow into manageable steps, enhancing the agent's ability to operate complex design software. BIMgent employs two execution workflows: a Pure-Action Workflow for common operations and a Vision-Driven Workflow for tasks requiring detailed GUI understanding. The framework's capabilities include generating 2D floorplans from design concepts, planning actions based on software documentation, and executing tasks autonomously while maintaining design intent. Evaluation results indicate that BIMgent produces reasonable design quality, surpasses baseline performance, and significantly reduces manual workload. The report further discusses related work and the challenges of applying autonomous agents to specialized building designs.