Research map / Multi-agent systems and games / Human-AI and goal-conditioned
THOMAS: Trajectory Heatmap Output with learned Multi-Agent Sampling
Thomas Gilles, Stefano Sabatini, Dzmitry Tsishkou, Bogdan Stanciulescu, Fabien Moutarde
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 learned scene-level trajectory recombination to marginal predictions (ablation) | THOMAS · Interaction validation | Joint MR: 14.8 -> 11.8 | Improved |
| add learned scene-level trajectory recombination to marginal predictions (ablation) | THOMAS · Interaction validation | Collision rate: 7.2% to 2.4% | Improved |
| replace learned recombination with collision-aware algorithmic sampling (ablation) | THOMAS · Interaction validation | Collision rate: 7.2 -> 2.6 | Improved |
| replace learned recombination with collision-aware algorithmic sampling (ablation) | THOMAS · Interaction validation | Joint MR: 14.8 -> 14.8 | No clear effect |
| use learned sampling instead of deterministic sampling followed by recombination (ablation) | THOMAS · Interaction validation | Joint MR: 11.8 -> 15.2 | Hurt |
| replace heatmap output with scalar regression and a joint objective (ablation) | THOMAS · Interaction validation | Marginal MR: 8.2 -> 16.2 | Hurt |
| use marginal rather than joint training objective with algorithmic sampling (ablation) | THOMAS · Interaction validation | Collision rate: 2.6 -> 7.2 | Hurt |
| use hierarchical heatmap refinement instead of GOHOME lane rasters (ablation) | THOMAS · Inference with 128 agents | inference time: 90 ms -> 31 ms | Improved |
| use hierarchical heatmap refinement instead of GOHOME lane rasters (ablation) | THOMAS · Interaction validation | marginal MissRate6 | No clear effect |
| use THOMAS instead of SceneTransformer | joint multi-agent prediction · Interaction validation | Joint MR: 15.7 -> 11.8 | Improved |
| use THOMAS instead of DenseTNT | joint multi-agent prediction · Interaction regular scene test | cSMR: 22.4 -> 25.2 | Mixed |
Later papers that compared with this method
3 later accepted papers report THOMAS as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Density-Adaptive Model Based on Motif Matrix for Multi-Agent Trajectory Prediction (CVPR 2024 · better on 0 of 1 shared result)
- FJMP: Factorized Joint Multi-Agent Motion Prediction Over Learned Directed Acyclic Interaction Graphs (CVPR 2023 · better on 0 of 1 shared result)
- Leveraging Future Relationship Reasoning for Vehicle Trajectory Prediction (ICLR 2023 · better on 1 of 1 shared result)
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
- SMART: Simultaneous Multi-Agent Recurrent Trajectory Prediction (ECCV 2020 · similarity 0.89)
- Multi-agent Long-term 3D Human Pose Forecasting via Interaction-aware Trajectory Conditioning (CVPR 2024 · similarity 0.88)
- MotionDiffuser: Controllable Multi-Agent Motion Prediction Using Diffusion (CVPR 2023 · similarity 0.88)
- ECTraj: Enhanced Consistency Training for Multi-Agent Trajectory Prediction (ECCV 2026 · similarity 0.87)
- Generative Trajectory Stitching through Diffusion Composition (NeurIPS 2025 · similarity 0.86)
- Unified Uncertainty-Aware Diffusion for Multi-Agent Trajectory Modeling (CVPR 2025 · similarity 0.85)
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.
