Research map / Vision-language and multimodal / Robot action and navigation
AffordGen: Generating Diverse Demonstrations for Generalizable Object Manipulation with Affordance Correspondence
Jiawei Zhang, Kaizhe Hu, Yingqian Huang, Yuanchen Ju, Zhengrong Xue, Huazhe Xu
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 |
|---|---|---|---|
| generate demonstrations over many unseen meshes via affordance correspondence instead of DemoGen single-object augmentation | visuomotor policy · Teapot Pouring (sim, unseen meshes) | success rate: 0.131 -> 0.553 | Improved |
| use AffordGen instead of CPGen mesh stretching | visuomotor policy · Mug Hanging (sim, unseen meshes) | success rate: 0.502 -> 0.707 | Improved |
| spread 1000 demos over 1000 meshes with 1 demo each instead of fewer meshes (ablation) | AffordGen · Teapot Pouring (sim, unseen meshes) | success rate: 0.553 -> 0.242 | Hurt |
| increase meshes from 20x50 to 50x20 for teapot (ablation) | AffordGen · Teapot Pouring (sim, unseen meshes) | success rate: 0.353 -> 0.553 | Improved |
| vary mesh count under fixed 1000 demos on shoe task (ablation) | AffordGen · Shoe Aligning (sim, unseen meshes) | success rate: 0.438 (20x50), 0.302 (50x20), 0.588 (100x10), 0.425 (1000x1) | Mixed |
| AffordGen training on generated data vs DemoGen on source mesh | visuomotor policy · Teapot Pouring (sim, source mesh) | success rate: 0.933 -> 0.960 | No clear effect |
| AffordGen vs CPGen on cross-category transfer | visuomotor policy · Teapot to Mug cross-category (sim) | success rate (%): 2.70 -> 55.00 | Improved |
| AffordGen closed-loop policy trained on generated data vs open-loop FUNCTO planner | real-world mug hanging · Mug Hanging (real, unseen objects) | successes/trials: 48/162 -> 107/162 | Improved |
| AffordGen vs CPGen on real knife cutting source object | visuomotor policy · Knife Cutting (real, source mesh) | successes/trials: 23/27 vs 23/27 | No clear effect |
| AffordGen vs CPGen on real shoe organizing source object | visuomotor policy · Shoe Organizing (real, source mesh) | successes/trials: 18/20 -> 11/20 | Hurt |
Closest papers by meaning
- AffordDP: Generalizable Diffusion Policy with Transferable Affordance (CVPR 2025 · similarity 0.92)
- Learning Precise Affordances from Egocentric Videos for Robotic Manipulation (ICCV 2025 · similarity 0.91)
- A0: An Affordance-Aware Hierarchical Model for General Robotic Manipulation (ICCV 2025 · similarity 0.90)
- G3Flow: Generative 3D Semantic Flow for Pose-aware and Generalizable Object Manipulation (CVPR 2025 · similarity 0.89)
- VidBot: Learning Generalizable 3D Actions from In-the-Wild 2D Human Videos for Zero-Shot Robotic Manipulation (CVPR 2025 · similarity 0.88)
- Robots Pre-train Robots: Manipulation-Centric Robotic Representation from Large-Scale Robot Datasets (ICLR 2025 · similarity 0.88)
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.
