MANSION: Multi-floor lANguage-to-3D Scene generatIOn for loNg-horizon tasks

TL;DR

MANSION introduces a novel framework for generating multi-floor 3D environments driven by language, enabling realistic, complex, and navigable building-scale tasks for advancing spatial reasoning in robotics.

Contribution

We present MANSION, the first language-driven system for creating multi-floor 3D environments, along with MansionWorld dataset and a scene editing agent for long-horizon tasks.

Findings

State-of-the-art agents perform poorly on MANSION environments.

MANSION provides diverse, realistic multi-floor scenes for benchmarking.

The framework supports customization of environments via open-vocabulary commands.

Abstract

Real-world robotic tasks are long-horizon and often span multiple floors, demanding rich spatial reasoning. However, existing embodied benchmarks are largely confined to single-floor in-house environments, failing to reflect the complexity of real-world tasks. We introduce MANSION, the first language-driven framework for generating building-scale, multi-floor 3D environments. Being aware of vertical structural constraints, MANSION generates realistic, navigable whole-building structures with diverse, human-friendly scenes, enabling the development and evaluation of cross-floor long-horizon tasks. Building on this framework, we release MansionWorld, a dataset of over 1,000 diverse buildings ranging from hospitals to offices, alongside a Task-Semantic Scene Editing Agent that customizes these environments using open-vocabulary commands to meet specific user needs. Benchmarking reveals that state-of-the-art agents degrade sharply in our settings, establishing MANSION as a critical testbed for the next generation of spatial reasoning and planning.