AIrchitect: Domain-Grounded Interactive 3D World Generation for Social Robotics

Volodymyr Shcherbyna, Do Duc Anh, Duc Tai Nguyen, Linh Kästner

Reviewer: ~Volodymyr_Shcherbyna1

Keywords: world generation, 3d world, photorealisic sim, indoor world

TL;DR: We present a modular platform that transforms user prompts to simulable 3D environments through a series of different world generation approaches.

Abstract:

Training and validating embodied AI requires realistic simulated indoor environments, yet existing generation methods are difficult to combine and rarely support converting real floorplans into simulation.
We introduce AIrchitect, an open-source platform that generates interactive 3D indoor environments from natural language prompts or floorplan images.
Its microservice architecture lets users swap between diffusion-based, procedural, and LLM-based backends for both floorplan generation and asset placement, all behind a common API.
A web-based frontend exposes the full pipeline from domain selection, through interactive editing, to export.
We generate 450 worlds across residential, hospital, and office domains and report structural and perceptual quality metrics.
The repository is publicly available at https://github.com/voshch/AIrchitect.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 26