Research map / Reinforcement learning / Visual and representation RL
DiffSkill: Skill Abstraction from Differentiable Physics for Deformable Object Manipulations with Tools
Xingyu Lin, Zhiao Huang, Yunzhu Li, Joshua B. Tenenbaum, David Held, 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 |
|---|---|---|---|
| compose learned skills with planning over tool choices and intermediate goals instead of using the strongest baseline | DiffSkill · LiftSpread | normalized improvement: 0.818 -> 0.920 | Improved |
| compose learned skills with planning over tool choices and intermediate goals instead of using the strongest baseline | DiffSkill · GatherTransport | normalized improvement: 0.567 -> 0.683 | Improved |
| compose learned skills with planning over tool choices and intermediate goals instead of using the strongest baseline | DiffSkill · CutRearrange | normalized improvement: 0.353 -> 0.297 | Hurt |
| add discrete planning over tool choices (ablation) | DiffSkill · LiftSpread | normalized improvement: 0.758 -> 0.920 | Improved |
| add discrete planning over tool choices (ablation) | DiffSkill · GatherTransport | normalized improvement: 0.312 -> 0.683 | Improved |
| add discrete planning over tool choices (ablation) | DiffSkill · CutRearrange | normalized improvement: 0.118 -> 0.297 | Improved |
| add intermediate-goal planning with randomly chosen skills replaced by planned tool choices (ablation) | DiffSkill · LiftSpread | normalized improvement: 0.593 -> 0.920 | Improved |
| add intermediate-goal planning with randomly chosen skills replaced by planned tool choices (ablation) | DiffSkill · GatherTransport | normalized improvement: 0.369 -> 0.683 | Improved |
| add intermediate-goal planning with randomly chosen skills replaced by planned tool choices (ablation) | DiffSkill · CutRearrange | normalized improvement: 0.018 -> 0.297 | Improved |
| add intermediate goals rather than directly supplying the final goal, even when the skill order is oracle-selected (ablation) | DiffSkill · LiftSpread | normalized improvement: 0.865 -> 0.920 | Improved |
| add intermediate goals rather than directly supplying the final goal, even when the skill order is oracle-selected (ablation) | DiffSkill · GatherTransport | normalized improvement: 0.501 -> 0.683 | Improved |
| add intermediate goals rather than directly supplying the final goal, even when the skill order is oracle-selected (ablation) | DiffSkill · CutRearrange | normalized improvement: 0.334 -> 0.297 | No clear effect |
Closest papers by meaning
- DiffVL: Scaling Up Soft Body Manipulation using Vision-Language Driven Differentiable Physics (NeurIPS 2023 · similarity 0.90)
- DeformGen: Dynamics-Based Topology Augmentation for Deformable Manipulation Policy Learning (NeurIPS 2026 · similarity 0.89)
- DexDeform: Dexterous Deformable Object Manipulation with Human Demonstrations and Differentiable Physics (ICLR 2023 · similarity 0.89)
- DiffTORI: Differentiable Trajectory Optimization for Deep Reinforcement and Imitation Learning (NeurIPS 2024 · similarity 0.89)
- Prof. Robot: Differentiable Robot Rendering Without Static and Self-Collisions (CVPR 2025 · similarity 0.86)
- Thin-Shell Object Manipulations With Differentiable Physics Simulations (ICLR 2024 · 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.
