Driving · 2026-10-26
SceneBloom: Agentic Indoor Environment Generation with Realistic and Diverse Layouts for Embodied Learning
Yuhan Liu, Lianhui Qin, Shenji Wan
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
SceneBloom generates simulation-ready 3D indoor environments for embodied learning. It samples furniture layouts per functional zone using diffusion priors, then compiles proposals into 3D with contract-guided fitting: relation contracts extracted from the proposals preserve qualitative relations while geometric fitting repairs poses. Across five room types, the authors report highest visual quality and physical stability, with no detected mesh collisions and no out-of-bounds objects.
Why it matters
The paper targets realistic, diverse, and simulation-ready indoor scenes for training embodied agents. SceneBloom’s two-stage approach separates qualitative spatial relations (preserved via relation contracts) from geometric pose repair (handled in 3D compilation), aiming to avoid collisions and maintain functional layouts. The authors further claim better transfer for object navigation on held-out real scans compared to policies trained on an equal number of real training scans.
Method
- Use a vision-language model to divide a room into functional zones and specify each zone’s furniture inventory and approximate floor region.
- For each zone, sample a visual furniture proposal with a text-to-image diffusion model conditioned on the zone prompt and asset-specific LoRA adapters, producing one coherent furniture group per proposal.
- Lift proposals into 3D with relation contracts extracted from the proposal images; a hierarchical compiler fits objects within zones, assembles zones into a room, and refines object poses to restore contract relations while maintaining collision-free feasibility.
Abstract (from arXiv)
Training generalizable embodied agents requires realistic, varied, and simulation-ready indoor environments. Language-guided methods support semantic planning and object-level reasoning, but precise spatial grounding remains difficult. Visual generative models offer rich priors over plausible arrangements, yet their outputs require geometric reasoning to become physically valid 3D scenes. We introduce SceneBloom, an agentic framework that combines visual layout sampling with contract-guided 3D compilation. A hierarchical sampler uses diffusion priors to generate furniture arrangements within functional zones. A compiler then uses relations extracted from each proposal to guide geometric fitting and local pose correction. Across five room types, SceneBloom attains the highest visual quality and physical stability among the compared methods and is the only method with no detected mesh collisions and out-of-bounds objects. Its layouts are also closer to the distribution of real rooms than those of all compared baselines. Object navigation policies trained in these realistic scenes achieve a higher success rate in held-out real scans than those trained on an equal number of real training scans. Code is available at https://github.com/Tinaliu0123/scenebloom.
Related papers
- AgentVQA: A Unified Benchmark for Agentic Visual Understanding
- Video2World: Benchmarking Coding Agents for Interactive World Modeling from Embodied Videos
- Awomo-SimDataEngine: Agentic Simulation-ReadyWorld Generation
- EmbodiedWorld-200K: Toward Embodied World Agents via Embodied-Planning Dataset and Interactive World Models
- Code2Worlds-2: Physically Plausible Agent Harness for Embodied World Simulation
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-23 · Indexed by Pub-AI: 2026-10-26 UTC
SceneBloom: Agentic Indoor Environment Generation with Realistic and Diverse Layouts for Embodied Learning
Preprint posted: 2026-10-23 · Original record ↗
- arxiv
- 2610.32978
Explicit source versions (1)
- v1 · 2026-10-23
- Preprint available
Source date: 2026-10-23 · First detected: 2026-10-27 UTC · Historical baseline
Source evidence (2)
- Official evidence ↗ · atom:published (official metadata)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
Training generalizable embodied agents requires realistic, varied, and simulation-ready indoor environments. Language-guided methods support semantic planning and object-level reasoning, but precise spatial grounding remains difficult. Visual generative models offer rich priors over plausible arrangements, yet their outputs require geometric reasoning to become physically valid 3D scenes. We introduce SceneBloom, an agentic framework that combines visual layout sampling with contract-guided 3D compilation. A hierarchical sampler uses diffusion priors to generate furniture arrangements within functional zones. A compiler then uses relations extracted from each proposal to guide geometric fitting and local pose correction. Across five room types, SceneBloom attains the highest visual quality and physical stability among the compared methods and is the only method with no detected mesh collisions and out-of-bounds objects. Its layouts are also closer to the distribution of real rooms than those of all compared baselines. Object navigation policies trained in these realistic scenes achieve a higher success rate in held-out real scans than those trained on an equal number of real training scans. Code is available at https://github.com/Tinaliu0123/scenebloom.
