Research map / Multi-agent systems and games / Human-AI and goal-conditioned
SMART: Simultaneous Multi-Agent Recurrent Trajectory Prediction
Sriram N N, Buyu Liu, Francesco Pittaluga, Manmohan Chandraker
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 pretraining on simulated P-ArgoT before fine-tuning on ArgoT | SMART (c_best) · ArgoT test 5.0s | ADE: 2.45 -> 2.02 | Improved |
| add pretraining on simulated P-ArgoT before fine-tuning on ArgoT | SMART (c_random) · ArgoT test 5.0s | ADE: 2.88 -> 2.39 | Improved |
| add pretraining on simulated P-ArgoT before fine-tuning on ArgoT | S-GAN · ArgoT test 5.0s | ADE: 4.06 -> 4.74 | Hurt |
| add pretraining on simulated P-ArgoT before fine-tuning on ArgoT | MATF GAN · ArgoT test 5.0s | ADE: 4.90 -> 5.07 | Hurt |
| add pretraining on simulated P-ArgoT | S-GAN · ArgoT test without straight trajectories 5.0s | ADE: 5.47 -> 4.73 | Improved |
| add pretraining on simulated P-ArgoT | MATF GAN · ArgoT test without straight trajectories 5.0s | ADE: 5.98 -> 5.61 | Improved |
| train on simulated P-ArgoT instead of ArgoT, test on KITTI | SMART (c_best) · KITTI 5.0s | ADE: 5.65 -> 5.32 | Improved |
| sample trajectory labels equally and take best instead of random labels | SMART · ArgoF val 3.0s | ADE: 1.65 -> 1.44 | Improved |
| sample trajectory labels equally and take best instead of random labels | SMART · ArgoF val 3.0s | FDE: 3.00 -> 2.47 | Improved |
| use single-pass ConvLSTM over all agents instead of per-agent prediction as agents grow | SMART vs S-GAN · Argoverse runtime | seconds per sample: .070 -> .069 for SMART vs .024 -> .108 for S-GAN from 1 to 10 agents | Improved |
| simulate diverse maneuvers instead of using real tracklets only | dataset · KITTI Y Wasserstein mean | diversity: 0.14 -> 2.13 | Improved |
| simulate diverse maneuvers instead of using real tracklets only | dataset · ArgoT second-derivative-x Wasserstein mean | diversity: 5.98 -> 17.5 | Improved |
Later papers that compared with this method
5 later accepted papers report SMART* as a baseline in their results tables; 5 of them report a better number on most of the results they share with it.
- Advancing Multi-agent Traffic Simulation via R1-Style Reinforcement Fine-Tuning (ICLR 2026 · better on 5 of 5 shared results)
- SPACeR: Self-Play Anchoring with Centralized Reference Models (ICLR 2026 · better on 3 of 4 shared results)
- Long-term Traffic Simulation via Structured Autoregressive Modeling (ECCV 2026 · better on 3 of 4 shared results)
- DecompGAIL: Learning Realistic Traffic Behaviors with Decomposed Multi-Agent Generative Adversarial Imitation Learning (ICLR 2026 · better on 2 of 3 shared results)
- Closed-Loop Supervised Fine-Tuning of Tokenized Traffic Models (CVPR 2025 · better on 3 of 3 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Diverse and Admissible Trajectory Prediction through Multimodal Context Understanding (ECCV 2020 · similarity 0.93)
- DeMo: Decoupling Motion Forecasting into Directional Intentions and Dynamic States (NeurIPS 2024 · similarity 0.89)
- HPNet: Dynamic Trajectory Forecasting with Historical Prediction Attention (CVPR 2024 · similarity 0.89)
- SmartRefine: A Scenario-Adaptive Refinement Framework for Efficient Motion Prediction (CVPR 2024 · similarity 0.89)
- ModeSeq: Taming Sparse Multimodal Motion Prediction with Sequential Mode Modeling (CVPR 2025 · similarity 0.88)
- Motion Forecasting in Continuous Driving (NeurIPS 2024 · 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.
