Research map / 3D vision / 3D generation
Hand-object interaction generation
Generates 3D hand-object and grasp motion with contact guidance and diffusion models.
116 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2019–2026) · an approach in 3D generation, keyword cluster “motion · interaction · human”
State History Tried Baselines Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
35 papers in 2023–24 and 75 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 7.7% to 8.5%.
Open questions the data points to
- Papers report MOHO on DexYCB (F-5) anywhere from 0.119 to 0.6: comparisons against it depend on whose number you take.
- Papers report MOHO on DexYCB (F-10) anywhere from 0.249 to 0.81: comparisons against it depend on whose number you take.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2019 · Learning Joint Reconstruction of Hands and Manipulated Objects (CVPR 2019 · 527 citations)
- 2022 · Ditto: Building Digital Twins of Articulated Objects From Interaction (CVPR 2022 · 89 citations)
- 2023 · ARCTIC: A Dataset for Dexterous Bimanual Hand-Object Manipulation (CVPR 2023 · 162 citations)
- 2024 · EMAGE: Towards Unified Holistic Co-Speech Gesture Generation via Expressive Masked Audio Gesture Modeling (CVPR 2024 · 79 citations)
- 2025 · GigaHands: A Massive Annotated Dataset of Bimanual Hand Activities (CVPR 2025 · 11 citations)
- 2026 · Towards Affordance-Aware Robotic Dexterous Grasping with Human-like Priors (AAAI 2026 · 2 citations)
Earliest papers here
- Learning Joint Reconstruction of Hands and Manipulated Objects (CVPR 2019 · 527 citations)
- ContactDB: Analyzing and Predicting Grasp Contact via Thermal Imaging (CVPR 2019 · 202 citations)
- Ditto: Building Digital Twins of Articulated Objects From Interaction (CVPR 2022 · 89 citations)
What has been tried
62 papers here report 594 measured changes (a component added, removed or swapped, compared with a baseline): helped 81%, no clear effect 3%, hurt 14%, mixed 2%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| add one VQ-VAE for full-body movements | BEATv1.3 · FGD | 13.080 -> 9.787 | Helped | EMAGE: Towards Unified Holistic Co-Speech Gesture Generation via Expressive Masked Audio Gesture Modeling (CVPR 2024 · 79 citations) |
| add human-object-contact cross-attention | BEHAVE and CHAIRS · R-Prec. (top-3) | 0.35 -> 0.62; 0.49 -> 0.74 | Helped | CG-HOI: Contact-Guided 3D Human-Object Interaction Generation (CVPR 2024 · 48 citations) |
| Combine guide-pose conditioning with audio conditioning in pose diffusion | Conversation test set · FDg | 3.94 -> 2.94 | Helped | From Audio to Photoreal Embodiment: Synthesizing Humans in Conversations (CVPR 2024 · 44 citations) |
| Add joint hand modeling to the object reconstruction | HO3D-v3 · CD [cm2] | 0.41 -> 0.38 | Helped | HOLD: Category-agnostic 3D Reconstruction of Interacting Hands and Objects from Video (CVPR 2024 · 37 citations) |
| add frame-wise and agent-wise positional encoding | GRAB · Physical realism | 0.8024 ± 0.0007 -> 0.8312 ± 0.0005 | Helped | Text2HOI: Text-guided 3D Motion Generation for Hand-Object Interaction (CVPR 2024 · 34 citations) |
| use an offset-vector interaction field rather than a scalar-distance field | Sitting · D2O ≤2cm (%) | 80.9 -> 99.6 | Helped | NIFTY: Neural Object Interaction Fields for Guided Human Motion Synthesis (CVPR 2024 · 31 citations) |
| Add global attention (GA) to the full model | PartNet-Mobility test set (K=32) · MMD (ID) | 0.053 -> 0.049 | Helped | CAGE: Controllable Articulation GEneration (CVPR 2024 · 22 citations) |
| add flow information to the input | t0 · success rate | 31.66 -> 41.43; 9.77% decrease without flow | Helped | GenH2R: Learning Generalizable Human-to-Robot Handover via Scalable Simulation Demonstration and Imitation (CVPR 2024 · 19 citations) |
| remove optimization and use the conditional denoised output directly | FlickrCI3D Signatures · joint PA-MPJPE (mm) | 66 -> 92 | Hurt | Generative Proxemics: A Prior for 3D Social Interaction from Images (CVPR 2024 · 19 citations) |
| Add interaction-environment inputs to the grasp synthesizer | S1 & S3 · Contact ratio (%) | 76.27 -> 82.86 | Helped | TACO: Benchmarking Generalizable Bimanual Tool-ACtion-Object Understanding (CVPR 2024 · 19 citations) |
| Jointly model hand and object rather than condition on hand pose | HOI4D · MPJPE | 1.14 -> 1.05 | Helped | G-HOP: Generative Hand-Object Prior for Interaction Reconstruction and Grasp Synthesis (CVPR 2024 · 18 citations) |
| Add estimated normal maps | DexYCB synthesis (1k images) · FID | 82.40 -> 77.64 | Helped | HOIDiffusion: Generating Realistic 3D Hand-Object Interaction Data (CVPR 2024 · 18 citations) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| MOHO | DexYCB | F-5 | 0.119 – 0.6 | 2 papers, e.g. CVPR 2025 and AAAI 2025 |
| MOHO | DexYCB | F-10 | 0.249 – 0.81 | 2 papers, e.g. CVPR 2025 and AAAI 2025 |
What reviewers will expect
Datasets and baselines named in the 33 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- GRAB · 5 of 33 papers
- ARCTIC · 5 of 33 papers
- BEHAVE · 4 of 33 papers
- HO3D · 3 of 33 papers
- DexYCB · 2 of 33 papers
- PartNet · 2 of 33 papers
- ShapeNet · 2 of 33 papers
- InterCap · 2 of 33 papers
Baselines
- MDM · 4 of 33 papers
- CHOIS · 4 of 33 papers
- OMOMO · 3 of 33 papers
- GraspTTA · 3 of 33 papers
- InterDiff · 3 of 33 papers
- PHOSA · 3 of 33 papers
- IMoS · 2 of 33 papers
- iHOI · 2 of 33 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 110 papers; from OpenAlex and the papers' affiliations).
- Tsinghua University · 8 papers
- Shanghai Jiao Tong University · 6 papers
- Shanghai Qi Zhi Institute · 5 papers
- Shanghai Artificial Intelligence Laboratory · 5 papers
- ShanghaiTech University · 5 papers
- Technical University of Munich · 5 papers
- NVIDIA · 4 papers
- Peking University · 4 papers
- Beijing University of Posts and Telecommunications · 3 papers
- Chinese Academy of Sciences · 3 papers
Use it
Download BibTeX of 30 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Monocular 3D object reconstruction (94 papers)
- Scene-aware human motion synthesis (82 papers)
- Learning physics from video (74 papers)
Elsewhere in the topic
- Learning 3D reconstruction from images (196 papers)
- Diffusion-based novel view synthesis (156 papers)
- Semantic scene completion (153 papers)
- Generative indoor scene synthesis (131 papers)
- Feed-forward 3D asset generation and editing (116 papers)
- Score distillation for text-to-3D (106 papers)
- Open-vocabulary 3D scene understanding (95 papers)
- 2D-to-3D pretrained representations (94 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).