Research map / Reinforcement learning / Visual and representation RL
Learning Foresightful Dense Visual Affordance for Deformable Object Manipulation
Ruihai Wu, Chuanruo Ning, Hao Dong
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 state value estimates (foresightful affordance) instead of greedy distance-only supervision (ablation) | dense affordance · SoftGym SpreadCloth | normalized score: 0.701 -> 0.758 | Improved |
| add state value estimates (ablation) | dense affordance · SoftGym RopeConfiguration | normalized score: 0.460 -> 0.529 | Improved |
| add value-aware stages beyond stage 1 (more stages of data) (ablation) | dense affordance · SoftGym SpreadCloth | normalized score: 0.589 -> 0.758 | Improved |
| use value-aware training at stage 1 versus greedy distance-only (ablation) | dense affordance · SoftGym SpreadCloth (stage 1) | normalized score: 0.701 -> 0.589 | Hurt |
| add Integrated Systematic Training (IST) (ablation) | dense affordance · SoftGym SpreadCloth | normalized score: 0.612 -> 0.758 | Improved |
| add IST (ablation) | dense affordance · DeformableRavens cable-ring | success rate (%): 78.3 -> 81.7 | Improved |
| use learned picking affordance instead of random picking (ablation) | dense affordance · DeformableRavens cable-ring | success rate (%): 11.7 -> 81.7 | Improved |
| use learned picking affordance instead of Transporter expert picking (ablation) | dense affordance · DeformableRavens cable-ring-notarget | success rate (%): 41.7 -> 95.0 | Improved |
| dense affordance learning without expert demonstrations versus Transporter imitation | Transporter · DeformableRavens cable-ring | success rate (%): 68.3 -> 81.7 | Improved |
| supervised dense affordance versus RL baselines | MVP, PlaNet, CURL-SAC, DrQ · SoftGym SpreadCloth | normalized score: 0.387 -> 0.758 | Improved |
| novel bead counts (generalization) versus Transporter | Transporter · cable-ring 24 beads | success rate (%): 33.3 -> 61.6 | Improved |
| real-world fine-tuned affordance versus MVP | MVP · Real world SpreadCloth | normalized score: 0.307 -> 0.683 | Improved |
Closest papers by meaning
- AffordDP: Generalizable Diffusion Policy with Transferable Affordance (CVPR 2025 · similarity 0.89)
- GarmentPile: Point-Level Visual Affordance Guided Retrieval and Adaptation for Cluttered Garments Manipulation (CVPR 2025 · similarity 0.88)
- DeformGen: Dynamics-Based Topology Augmentation for Deformable Manipulation Policy Learning (NeurIPS 2026 · similarity 0.88)
- IAAO: Interactive Affordance Learning for Articulated Objects in 3D Environments (CVPR 2025 · similarity 0.87)
- Entity-Centric Reinforcement Learning for Object Manipulation from Pixels (ICLR 2024 · similarity 0.87)
- DeformMaster: An Interactive Physics-Neural World Model for Deformable Objects from Videos (NeurIPS 2026 · 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.
