RoboBIM-Sim: Semantically Articulated USD Environments from IFC Building Models
Overview
- Embodied AI requires large-scale, semantically rich simulation environments to train robust navigation and manipulation policies.
- Procedural generation and scanned datasets (e.g., Habitat-Matterport ) are common, but lack the exact parametric data of real-world buildings.
- Building Information Modeling (BIM) standards such as Industry Foundation Classes (IFC) contain exactly this ground-truth architectural data along with rich object semantics.
- However, IFC models lack the physics, rendering materials, and joint articulations required by modern robotics simulators such as NVIDIA Isaac Sim, which relies on Universal Scene Description (USD).
- This thesis bridges architectural BIM and Embodied AI: an automated pipeline ingests IFC building models and outputs semantically annotated, physics-ready USD scenes.
- By mapping IFC property sets to USD physics and articulation properties (hinge joints for doors, prismatic joints for drawers, based on IFC classifications), robots can be injected with a prior spatial memory of a building’s exact layout and interactive elements before deployment.
Objectives & Methodology
- Pipeline development: Create a Python/C++ toolchain to parse IFC hierarchies and translate them into USD format.
- Automated articulation: Extract semantic tags from IFC to automatically rig structural elements (doors, windows, cabinets) with correct physical joints in Isaac Sim.
- Validation: Demonstrate a metric-semantic navigation and basic manipulation task in Isaac Sim using the injected architectural prior.
Skills
- Python programming experience; C++ is a plus
- Familiarity with 3D scene representations and geometry processing
- Interest in robotics simulation (NVIDIA Isaac Sim / Isaac Lab, USD)
- Basic knowledge of BIM/IFC is a plus but can be acquired during the thesis
Tasks
- Literature research on BIM-to-simulation conversion and Embodied AI environments
- Implementation of the IFC parsing and USD export toolchain
- Automated joint/articulation rigging from IFC semantics
- Evaluation of navigation and manipulation tasks in Isaac Sim
Core References
- Makoviychuk, V., et al. (2021). “Isaac Gym: High Performance GPU-Based Physics Simulation For Robot Learning.” NeurIPS Datasets and Benchmarks Track. — Baseline for USD/physics simulation.
- Savva, M., et al. (2019). “Habitat: A Platform for Embodied AI Research.” ICCV. — Context for Embodied AI simulation needs.
- Krijnen, T. F., & Beetz, J. (2017). “An IFC Open Source engine for 3D visualization, semantic querying and generation of building geometries.” W78 Conference. — IFC parsing SOTA.
Contact
Ashwin Nedungadi