Research map / 3D vision / 3D generation
PhyScensis: Physics-Augmented LLM Agents for Complex Physical Scene Arrangement
Yian Wang, Han Yang, Minghao Guo, Xiaowen Qiu, Tsun-Hsuan Wang, Wojciech Matusik, Joshua B. Tenenbaum, Chuang Gan
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add the feedback system (vs binary success/failure only) (ablation) | PhyScensis · Retry Times | retries: 1.69 -> 1.04 | Improved |
| add the feedback system (ablation) | PhyScensis · Time-cost | time: 132.29 -> 106.41 | Improved |
| add empty-region reporting to feedback (ablation) | PhyScensis · Time-cost | time: 126.09 -> 106.41 | Improved |
| add empty-region reporting to feedback (ablation) | PhyScensis · Retry Times | retries: 1.43 -> 1.04 | Improved |
| add visual feedback to the full framework (ablation) | PhyScensis · Retry Times | retries: 1.04 -> 0.95 | Mixed |
| add designed predicates and solver instead of Random placement (ablation) | PhyScensis · tabletop scenes | VQA Score: 0.415 -> 0.704 | Improved |
| add designed predicates and solver instead of LLM-Only placement (ablation) | PhyScensis · tabletop scenes | VQA Score: 0.592 -> 0.704 | Improved |
| add solver and physics instead of LLM-Only placement (ablation) | PhyScensis · tabletop scenes | Settle Distance: 0.154 -> 0.003 | Improved |
| use PhyScensis instead of 3D-Generalist | scene generation · tabletop scenes | VQA Score: 0.578 -> 0.704 | Improved |
| use PhyScensis scenes for imitation data instead of 3D-Generalist | diffusion policy · robot reaching/placing | success rate: 4/10 -> 9/10 reaching; 1/10 -> 3/10 placing | Improved |
| use PhyScensis instead of LayoutVLM | scene generation · LayoutVLM comparison | Settle Distance: 0.115 -> 0.003 | Improved |
| use PhyScensis instead of ClutterGen | object arrangement · 500 sampled scenes | objects placed: 3.57 -> 14.77 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Architect — Architect: Generating Vivid and Interactive 3D Scenes with Hierarchical 2D Inpainting (NeurIPS 2024) · better on 8 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- PhyScene3D: Physically Consistent 3D Interactive Tabletop Scene Generation (ICML 2026 · similarity 0.90)
- PhyScene: Physically Interactable 3D Scene Synthesis for Embodied AI (CVPR 2024 · similarity 0.89)
- SceneSmith: Agentic Generation of Simulation-Ready Indoor Scenes (ICML 2026 · similarity 0.89)
- PhysX-3D: Physical-Grounded 3D Asset Generation (NeurIPS 2025 · similarity 0.88)
- Simulation-Ready Compositional 3D Scene Reconstruction from a Single Image (NeurIPS 2026 · similarity 0.87)
- SceneWeaver: All-in-One 3D Scene Synthesis with an Extensible and Self-Reflective Agent (NeurIPS 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
