Research map / 3D vision / Human pose estimation
PartManip: Learning Cross-Category Generalizable Part Manipulation Policy From Point Cloud Observations
Haoran Geng, Ziming Li, Yiran Geng, Jiayi Chen, Hao Dong, He Wang
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 part-pose canonicalization to the state-based expert (ablation) | PartManip state-based expert · OpenDoor, Training Set | success rate (%): 67.8±3.4 -> 82.2±0.2 | Improved |
| Add part-pose canonicalization to the state-based expert (ablation) | PartManip state-based expert · OpenDoor, Val-Inter Set | success rate (%): 23.4±3.9 -> 50.7±4.1 | Improved |
| Use DAgger instead of directly training with PPO (ablation) | PartManip vision policy · OpenDoor, Val-Inter Set | success rate (%): 0.2±0.2 -> 31.5±2.9 | Improved |
| Add point-cloud augmentation and Sparse-UNet to DAgger (ablation) | PartManip vision policy · OpenDoor, Val-Inter Set | success rate (%): 31.5±2.9 -> 41.7±2.5 | Improved |
| Add expert-demonstration pretraining (ablation) | PartManip vision policy with DAgger, augmentation, Sparse-UNet · OpenDoor, Val-Inter Set | success rate (%): 41.7±2.5 -> 37.2±1.2 | Hurt |
| Add domain-adversarial learning (ablation) | PartManip vision policy with DAgger, augmentation, Sparse-UNet, pretraining · OpenDoor, Val-Inter Set | success rate (%): 37.2±1.2 -> 49.1±1.5 | Improved |
| Add domain-adversarial learning (ablation) | PartManip vision policy with DAgger, augmentation, Sparse-UNet, pretraining · OpenDrawer, Val-Inter Set | success rate (%): 48.8±4.5 -> 54.7±4.2 | Improved |
| Use PartManip rather than the strongest listed baseline | OpenDoor benchmark · OpenDoor Val-Inter Set | success rate (%): 26.9±2.2 -> 49.1±1.5 | Improved |
| Use PartManip rather than the strongest listed baseline | OpenDrawer benchmark · OpenDrawer Val-Inter Set | success rate (%): 31.2±5.1 -> 54.7±4.2 | Improved |
Closest papers by meaning
- Entity-Centric Reinforcement Learning for Object Manipulation from Pixels (ICLR 2024 · similarity 0.87)
- AffordDP: Generalizable Diffusion Policy with Transferable Affordance (CVPR 2025 · similarity 0.87)
- EC-Diffuser: Multi-Object Manipulation via Entity-Centric Behavior Generation (ICLR 2025 · similarity 0.86)
- PartRM: Modeling Part-Level Dynamics with Large Cross-State Reconstruction Model (CVPR 2025 · similarity 0.86)
- General Articulated Objects Manipulation in Real Images via Part-Aware Diffusion Process (NeurIPS 2024 · similarity 0.86)
- PointMapPolicy: Structured Point Cloud Processing for Multi-Modal Imitation Learning (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.
