Research map / Probabilistic and scientific modeling / Dynamical systems
Contact-Aware Neural Dynamics
Changwei Jing, Jai Krishna Bandi, Jianglong Ye, Yan Duan, Pieter Abbeel, Xiaolong Wang, Sha Yi
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 binary contact prediction and conditioning to Diffusion-UNet (ablation) | Diffusion-UNet · single object, real fine-tune | ADD-S AUC (%): 82.45 -> 88.23 | Improved |
| add binary contact prediction and conditioning to Diffusion-UNet (ablation) | Diffusion-UNet · single object, real fine-tune | MSE: 0.0091 -> 0.0082 | Improved |
| add binary contact prediction and conditioning to Diffusion-UNet (ablation) | Diffusion-UNet · multiple objects, real fine-tune | ADD-S AUC (%): 75.82 -> 79.12 | Improved |
| add binary contact prediction and conditioning to Diffusion-UNet (ablation) | Diffusion-UNet · single object, sim data | ADD-S AUC (%): 69.12 -> 68.12 | Hurt |
| add binary contact prediction and conditioning to Diffusion-UNet (ablation) | Diffusion-UNet · single object, real data | ADD-S AUC (%): 80.03 -> 81.34 | Improved |
| add binary contact prediction and conditioning to Diffusion-UNet (ablation) | Diffusion-UNet · multiple objects, real data | MSE: 0.0083 -> 0.0075 | Improved |
| replace MLP with Diffusion-UNet dynamics head (ablation) | MLP · single object, real data | ADD-S AUC (%): 78.12 -> 80.03 | Improved |
| replace plain UNet with diffusion UNet (ablation) | UNet · single object, real data | ADD-S AUC (%): 68.45 -> 80.03 | Improved |
| fine-tune on real data after simulation pretraining | Diffusion-UNet w/ Contact · single object | ADD-S AUC (%): 81.34 -> 88.23 | Improved |
| fine-tune on real data after simulation pretraining | MLP · single object | ADD-S AUC (%): 78.12 -> 77.43 | Hurt |
| pretrain on simulation then fine-tune on real data instead of real-only training | contact-aware dynamics · single-object task success | success rate (%): 52.6 -> 73.7 | Improved |
| pretrain on simulation then fine-tune on real data instead of real-only training | contact-aware dynamics · multi-object task success | success rate (%): 47.1 -> 64.7 | Improved |
Closest papers by meaning
- Enhancing Tactile-based Reinforcement Learning for Robotic Control (NeurIPS 2025 · similarity 0.88)
- DIFFTACTILE: A Physics-based Differentiable Tactile Simulator for Contact-rich Robotic Manipulation (ICLR 2024 · similarity 0.87)
- ForceFlow: Learning to Feel and Act via Contact-Driven Flow Matching (NeurIPS 2026 · similarity 0.87)
- Geometric Contact Flows: Contactomorphisms for Dynamics and Control (ICML 2025 · similarity 0.86)
- EgoTac: In-the-wild Tactile Prediction from Egocentric Vision (NeurIPS 2026 · similarity 0.86)
- Rapidly Adapting Policies to the Real-World via Simulation-Guided Fine-Tuning (ICLR 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.
