Research map / 3D vision / 3D generation
3D-aware Disentangled Representation for Compositional Reinforcement Learning
Sungbin Mun, Younghwan Lee, Cheol-Hui Min, Mineui Hong, Young Min Kim
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 |
|---|---|---|---|
| use 3D block-slot attention with factor binding instead of permutation-invariant OSRT slots | OSRT · Clevr3D | FG-ARI: 0.365 -> 0.942 | Improved |
| use 3D block-slot attention instead of OSRT slots | OSRT · IsaacGym3D | FG-ARI: 0.321 -> 0.619 | Improved |
| use 3D block-slot attention instead of OSRT slots | OSRT · IsaacGym3D | PSNR: 27.35 -> 26.55 | No clear effect |
| use 3D block-slot attention instead of OSRT slots | OSRT · Clevr3D | DCI informativeness: 0.452 -> 0.844 | Improved |
| replace entity-wise EIT policy with block transformer on block-slot representation (ablation) | Ours w/ EIT · CG (same color) | success rate: 0.682 -> 0.837 | Improved |
| replace entity-wise EIT policy with block transformer (ablation) | Ours w/ EIT · CG | success rate: 0.773 -> 0.895 | Improved |
| replace entity-wise EIT policy with block transformer (ablation) | Ours w/ EIT · ID | success rate: 0.984 -> 0.967 | No clear effect |
| replace 2D DLPv2 representation with 3D block-slot representation (EIT policy) | DLPv2 w/ EIT · CG (same color) | success rate: 0.388 -> 0.682 | Improved |
| evaluate on unseen viewpoints, 2D baseline vs 3D block-slot with BT | DLPv2 w/ EIT · OOD multi-view ID goals | success rate: 0.059 -> 0.948 | Improved |
| use single view at inference instead of three views (ablation) | 3D block-slot representation · IsaacGym3D | PSNR: 26.55 -> 25.90 | Hurt |
| use object-centric representation instead of VAE with MLP policy | VAE w/ MLP · IsaacGym push | success rate: 0.042 -> 0.967 | Improved |
Closest papers by meaning
- Entity-Centric Reinforcement Learning for Object Manipulation from Pixels (ICLR 2024 · similarity 0.87)
- MetaSpatial: Reinforcing 3D Spatial Reasoning in VLMs for the Metaverse (ICLR 2026 · similarity 0.86)
- Localizing, Structuring, and Rendering: Bridging 3D and 2D Vision-Language-Action Models for Robotic Manipulation (CVPR 2026 · similarity 0.86)
- SOLD: Slot Object-Centric Latent Dynamics Models for Relational Manipulation Learning from Pixels (ICML 2025 · similarity 0.85)
- Object centric architectures enable efficient causal representation learning (ICLR 2024 · similarity 0.85)
- Object-X: Learning to Reconstruct Multi-Modal 3D Object Representations (NeurIPS 2025 · similarity 0.85)
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.
