Research map / Probabilistic and scientific modeling / Molecules and proteins
Multimodal trajectory prediction
Predicts agent trajectories using causal disentanglement, trajectory sets and graph models.
95 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Molecules and proteins, keyword cluster “flow · trajectory · matching”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Few recent reports None of the tracked benchmarks has a report from 2025–26.
17 papers in 2023–24 and 47 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 3.5% to 5.1%.
No longer reported by papers here: ETH/UCY AVG (ADE, last in 2024); ETH/UCY AVG (FDE, last in 2022).
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
- 2016 · Minimizing User Involvement for Learning Human Mobility Patterns from Location Traces (AAAI 2016 · 11 citations)
- 2017 · Learning to Predict Intent from Gaze During Robotic Hand-Eye Coordination (AAAI 2017 · 20 citations)
- 2018 · Social GAN: Socially Acceptable Trajectories With Generative Adversarial Networks (CVPR 2018 · 2,462 citations)
- 2019 · Which Way Are You Going? Imitative Decision Learning for Path Forecasting in Dynamic Scenes (CVPR 2019 · 70 citations)
- 2020 · CoverNet: Multimodal Behavior Prediction Using Trajectory Sets (CVPR 2020 · 443 citations)
- 2021 · Trajectory Prediction With Latent Belief Energy-Based Model (CVPR 2021 · 90 citations)
- 2022 · Stochastic Trajectory Prediction via Motion Indeterminacy Diffusion (CVPR 2022 · 302 citations)
- 2023 · WSiP: Wave Superposition Inspired Pooling for Dynamic Interactions-Aware Trajectory Prediction (AAAI 2023 · 47 citations)
- 2024 · BAT: Behavior-Aware Human-Like Trajectory Prediction for Autonomous Driving (AAAI 2024 · 64 citations)
- 2025 · MoFlow: One-Step Flow Matching for Human Trajectory Forecasting via Implicit Maximum Likelihood Estimation based Distillation (CVPR 2025 · 20 citations)
- 2026 · DiffRefiner: Coarse to Fine Trajectory Planning via Diffusion Refinement with Semantic Interaction for End to End Autonomous Driving (AAAI 2026 · 2 citations)
Earliest papers here
- Minimizing User Involvement for Learning Human Mobility Patterns from Location Traces (AAAI 2016 · 11 citations)
- Learning to Predict Intent from Gaze During Robotic Hand-Eye Coordination (AAAI 2017 · 20 citations)
- Social GAN: Socially Acceptable Trajectories With Generative Adversarial Networks (CVPR 2018 · 2,462 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
ETH/UCY AVG · ADE (lower is better)
5 papers, 2021–2024 · not reported since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 0.21 | Trajectory Prediction With Latent Belief Energy-Based Model (CVPR 2021 · 90 citations) |
| 2024 | 0.03 | SocialCVAE: Predicting Pedestrian Trajectory via Interaction Conditioned Latents (AAAI 2024 · 10 citations) |
ETH/UCY AVG · FDE (lower is better)
4 papers, 2021–2022 · not reported since 2022
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 0.38 | Trajectory Prediction With Latent Belief Energy-Based Model (CVPR 2021 · 90 citations) |
| 2022 | 0.36 | Non-Probability Sampling Network for Stochastic Human Trajectory Prediction (CVPR 2022 · 77 citations) |
What has been tried
59 papers here report 604 measured changes (a component added, removed or swapped, compared with a baseline): helped 76%, no clear effect 7%, hurt 15%, 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 polar-coordinate representation instead of Cartesian coordinates | NGSIM overall test, 5 s · RMSE (m) | 4.02 -> 3.62 | Helped | BAT: Behavior-Aware Human-Like Trajectory Prediction for Autonomous Driving (AAAI 2024 · 64 citations) |
| add causal disentanglement to the DyHIN baseline | AGV2 validation · minADE6 | 0.93 -> 0.88 | Helped | CaDeT: a Causal Disentanglement Approach for Robust Trajectory Prediction in Autonomous Driving (CVPR 2024 · 26 citations) |
| Replace IMLE distillation with WGAN | NBA · min20ADE | 0.71 -> 1.78 | Hurt | MoFlow: One-Step Flow Matching for Human Trajectory Forecasting via Implicit Maximum Likelihood Estimation based Distillation (CVPR 2025 · 20 citations) |
| add social physics guidance and the crowd interaction module | GC · MAE | 3.3102 -> 0.9116 | Helped | Social Physics Informed Diffusion Model for Crowd Simulation (AAAI 2024 · 16 citations) |
| add semantic- and motion-conditional normalization in memory (ego-motion) | nuScenes Inflated GMO ablation · mIoU_f (1s) | +1.9 (26.4 to 28.3) | Helped | Driving in the Occupancy World: Vision-Centric 4D Occupancy Forecasting and Planning via World Models for Autonomous Driving (AAAI 2025 · 14 citations) |
| Add flow and object modalities to RGB | EK100 validation, K700 features · Top-5 recall (action) | 20.3 -> 23.0 | Helped | Uncertainty-aware Action Decoupling Transformer for Action Anticipation (CVPR 2024 · 14 citations) |
| add the attention model to the coarse prediction model | SDD · ADE | 8.64 -> 1.48 | Helped | SocialCVAE: Predicting Pedestrian Trajectory via Interaction Conditioned Latents (AAAI 2024 · 10 citations) |
| replace parallel DETR-style decoding with sequential mode decoding | WOMD validation, 20% training data · Soft mAP6 | 0.4050 -> 0.4138 | Helped | ModeSeq: Taming Sparse Multimodal Motion Prediction with Sequential Mode Modeling (CVPR 2025 · 9 citations) |
| Add the refinement module to C2F-TP (C) | NGSIM, 5s average · ADE | 0.81 -> 0.79 | Helped | C2F-TP: A Coarse-to-Fine Denoising Framework for Uncertainty-Aware Trajectory Prediction (AAAI 2025 · 8 citations) |
| Add the ground-truth trajectory loss LT alongside EmLoco loss | JTA · ADE | 3.52 -> 0.97 | Helped | Physical Plausibility-aware Trajectory Prediction via Locomotion Embodiment (CVPR 2025 · 8 citations) |
| Add the kinematics feature extractor | Argoverse · ECE | 0.1395 -> 0.1335 | Helped | CCTR: Calibrating Trajectory Prediction for Uncertainty-Aware Motion Planning in Autonomous Driving (AAAI 2024 · 8 citations) |
| add SVA | nuScenes val, generated samples · NDS | 32.1 -> 34.3 | Helped | SubjectDrive: Scaling Generative Data in Autonomous Driving via Subject Control (AAAI 2025 · 7 citations) |
What reviewers will expect
Datasets and baselines named in the 21 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- nuScenes · 5 of 21 papers
- highD · 2 of 21 papers
- ETH-UCY · 2 of 21 papers
- Waymo Open Motion Dataset · 2 of 21 papers
- Argoverse2 · 2 of 21 papers
- NAVSIM · 2 of 21 papers
- nuPlan · 2 of 21 papers
Baselines
- CS-LSTM · 2 of 21 papers
- LED · 2 of 21 papers
- UniAD · 2 of 21 papers
- VAD-Base · 2 of 21 papers
- Trajectron++ · 2 of 21 papers
- EqMotion · 2 of 21 papers
- QCNet · 2 of 21 papers
- MTR++ · 2 of 21 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 64 papers; from OpenAlex and the papers' affiliations).
- Tsinghua University · 4 papers
- Zhejiang University · 4 papers
- Huawei · 3 papers
- The Chinese University of Hong Kong · 3 papers
- Central South University · 2 papers
- Nanyang Technological University · 2 papers
- City University of Hong Kong · 2 papers
- Massachusetts Institute of Technology · 2 papers
- Peng Cheng Laboratory · 2 papers
- University College London · 2 papers
Use it
Download BibTeX of 39 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
- Flow matching generative modeling (97 papers)
- Spatiotemporal forecasting and imputation (30 papers)
Elsewhere in the topic
- Generative protein structure design (142 papers)
- Prediction-powered and conformal inference (129 papers)
- Statistical analysis of ML practice (122 papers)
- Neuro-symbolic probabilistic layers (120 papers)
- Structure-based 3D molecule generation (107 papers)
- Sparse graphical and tensor models (97 papers)
- Machine learning for physical simulation (96 papers)
- Probabilistic embeddings and relational inference (93 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).