Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Sim, Yet Same: Physics-Aligned Simulator as Zero-Shot Data Scaler in Deformable Worlds
Yunsong Zhou, Hangxu Liu, Xuekun Jiang, Xing Shen, Yuanzhen Zhou, Hui Wang, Baole Fang, Yang Tian, Mulin Yu, Qiaojun Yu, Li Ma, Hengjie Li and 3 more
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 trajectory decomposition to the MimicGen-style baseline (ablation) | SIM1 data generation · T-shirt folding | pass rate / success rate: pass rate 0% -> 65%, success 0% | Mixed |
| add diffusion-based motion generation (ablation) | SIM1 data generation · T-shirt folding in-domain | success rate (%): 0 -> 47 | Improved |
| add diffusion-based motion generation (ablation) | SIM1 data generation · T-shirt folding average | average success rate (%): 0 -> 33 | Improved |
| add deformation-stable solver (ablation) | SIM1 data generation · T-shirt folding in-domain | success rate (%): 47 -> 67 | Improved |
| add deformation-stable solver (ablation) | SIM1 data generation · T-shirt folding average | average success rate (%): 33 -> 76 | Improved |
| add deformation-stable solver (ablation) | SIM1 data generation · T-shirt folding viewpoint shift | success rate (%): 7 -> 47 | Improved |
| add deformation-stable solver (ablation) | SIM1 data generation · T-shirt folding pass rate | pass rate (%): 38 -> 40 | No clear effect |
| train on simulation-teleoperated data instead of real data | pi0.5 · T-shirt folding in-domain | average success rate (%): 97 -> 87 | Hurt |
| use simulation-generated data instead of real data under domain shift | pi0.5 · spatial, texture, lighting shifts | success rate gain (points): +50%, +13%, +47% | Improved |
| use synthetic data instead of real data when training from scratch | pi0.5 (from scratch) · T-shirt folding | success rate (%): 0% -> 76% | Improved |
| use synthetic data instead of real data on an unseen polo shirt | pi0.5 · real-robot polo shirt | success rate (%): 20% -> 70% | Improved |
| scale synthetic data volume | pi0.5 and pi0 · T-shirt folding in-domain | success rate: 15 synthetic samples per real demo; 5 for texture generalization | Improved |
Closest papers by meaning
- DeformMaster: An Interactive Physics-Neural World Model for Deformable Objects from Videos (NeurIPS 2026 · similarity 0.89)
- Simulation-Ready Compositional 3D Scene Reconstruction from a Single Image (NeurIPS 2026 · similarity 0.88)
- DeformGen: Dynamics-Based Topology Augmentation for Deformable Manipulation Policy Learning (NeurIPS 2026 · similarity 0.88)
- RoboScape: Physics-informed Embodied World Model (NeurIPS 2025 · similarity 0.88)
- InterMimic: Towards Universal Whole-Body Control for Physics-Based Human-Object Interactions (CVPR 2025 · similarity 0.86)
- PhysX-3D: Physical-Grounded 3D Asset Generation (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.
