Research map / 3D vision / Human pose estimation
Human motion prediction
Forecasts future 3D human motion from past observations using dynamic relationship modeling.
38 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2017–2026) · an approach in Human pose estimation, keyword cluster “body · motion · human mesh”
State History Tried 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.
11 papers in 2023–24 and 11 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 4.1% to 3.7%.
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
- 2017 · Forecasting Human Dynamics From Static Images (CVPR 2017 · 117 citations)
- 2019 · BiHMP-GAN: Bidirectional 3D Human Motion Prediction GAN (AAAI 2019 · 133 citations)
- 2020 · Learning Dynamic Relationships for 3D Human Motion Prediction (CVPR 2020 · 171 citations)
- 2021 · We Are More Than Our Joints: Predicting How 3D Bodies Move (CVPR 2021 · 128 citations)
- 2022 · Progressively Generating Better Initial Guesses Towards Next Stages for High-Quality Human Motion Prediction (CVPR 2022 · 160 citations)
- 2023 · Trajectory-Aware Body Interaction Transformer for Multi-Person Pose Forecasting (CVPR 2023 · 30 citations)
- 2024 · WANDR: Intention-guided Human Motion Generation (CVPR 2024 · 13 citations)
- 2025 · BEDLAM2.0: Synthetic humans and cameras in motion (NeurIPS 2025 · 7 citations)
- 2026 · Hi-EF: Benchmarking Emotion Forecasting in Human-interaction (AAAI 2026 · 1 citation)
Earliest papers here
- Forecasting Human Dynamics From Static Images (CVPR 2017 · 117 citations)
- BiHMP-GAN: Bidirectional 3D Human Motion Prediction GAN (AAAI 2019 · 133 citations)
- Human Motion Prediction via Learning Local Structure Representations and Temporal Dependencies (AAAI 2019 · 102 citations)
What has been tried
30 papers here report 323 measured changes (a component added, removed or swapped, compared with a baseline): helped 76%, no clear effect 4%, hurt 16%, mixed 4%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| train on both AMASS and CIRCLE instead of one dataset | AMASS+CIRCLE eval · Success Rate | 32% vs 16% (AMASS only) and 0% (CIRCLE only) | Helped | WANDR: Intention-guided Human Motion Generation (CVPR 2024 · 13 citations) |
| add TIA | GIMO · Traj-dest | 850 -> 666 | Helped | Multimodal Sense-Informed Forecasting of 3D Human Motions (CVPR 2024 · 8 citations) |
| add cross neutralization (CN) | GRAB, all actions, whole body · MPJPE (mm) | 64.4 -> 61.9 | Helped | Expressive Forecasting of 3D Whole-Body Human Motions (AAAI 2024 · 7 citations) |
| add cross-context interaction between body and hands | GRAB, whole body · MPJPE (mm) | 68.2 -> 65.4 | Helped | Forecasting of 3D Whole-body Human Poses with Grasping Objects (CVPR 2024 · 7 citations) |
| replace GCN with SAGGB | H3.6M average, 1000ms · MPJPE | 111.8 -> 110.4 | Helped | Existence Is Chaos: Enhancing 3D Human Motion Prediction with Uncertainty Consideration (AAAI 2024 · 5 citations) |
| Use characteristic poses rather than poses sampled independently of actions | MPII Cooking II · MPJPE [px] | 75 -> 50 | Helped | FutureHuman3D: Forecasting Complex Long-Term 3D Human Behavior from Video Observations (CVPR 2024 · 4 citations) |
| Add low-rank modulation | Human3.6M · RunTime | 38.73 -> 29.10 | Helped | ALIEN: Implicit Neural Representations for Human Motion Prediction under Arbitrary Latency (CVPR 2025 · 3 citations) |
| Increase latency length from TL=3 to TL=5 | Human3.6M average, 1000 ms · MPJPE | 111.7 -> 108.5 | Helped | LAL: Enhancing 3D Human Motion Prediction with Latency-aware Auxiliary Learning (CVPR 2025 · 1 citation) |
| Use pose only for the primary agent rather than all agents | JTA · ADE | 0.89 -> 0.92 | Hurt | Social-Transmotion: Promptable Human Trajectory Prediction (ICLR 2024 · 1 citation) |
What reviewers will expect
Datasets and baselines named in the 11 papers of this approach from 2024–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- Human3.6M · 5 of 11 papers
- 3DPW · 4 of 11 papers
- AMASS · 4 of 11 papers
- CMU Mocap · 3 of 11 papers
- GRAB · 3 of 11 papers
- ETH-UCY · 2 of 11 papers
Baselines
- LTD · 6 of 11 papers
- SPGSN · 6 of 11 papers
- PGBIG · 5 of 11 papers
- DMGNN · 4 of 11 papers
- EqMotion · 2 of 11 papers
- Res-sup · 2 of 11 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 22 papers; from OpenAlex and the papers' affiliations).
- Nanjing University of Science and Technology · 6 papers
- Zhejiang University · 3 papers
- Xiaohongshu Inc · 2 papers
- Max Planck Institute for Intelligent Systems · 2 papers
- ETH Zurich · 2 papers
Use it
Download BibTeX of 36 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
- Physics and sensor-based motion capture (125 papers)
- Human shape and pose regression (80 papers)
- Human mesh recovery from images (58 papers)
Elsewhere in the topic
- Learned camera pose estimation (96 papers)
- Category-level object pose estimation (92 papers)
- Monocular 2D-to-3D pose lifting (90 papers)
- Depth-based hand pose regression (84 papers)
- Transformer-based multi-view 3D pose (83 papers)
- Temporal video pose estimation (76 papers)
- Instance-level 6D object pose (74 papers)
- Category-level pose estimation (54 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).