Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Heteroscedastic Diffusion for Multi-Agent Trajectory Modeling
Guillem Capellera, Antonio Rubio, Luis Ferraz, Antonio Agudo
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 Temporal Mamba in place of Transformer encoder for temporal modeling (ablation) | U2Diff · NBA | minSADE20: 1.73 -> 1.47 | Improved |
| add Social Transformer for agent interactions (ablation) | U2Diff · Football-U | minSADE20: 4.75 -> 2.35 | Improved |
| add NLL loss term (lambda 0.01) (ablation) | U2Diff · Basketball-U and NBA | minSADE20: 3.10 -> 3.13 (Basketball-U); 1.50 -> 1.48 (NBA) | Mixed |
| propagate uncertainty with diagonal Jacobian instead of J_s=0 (ablation) | uni-variate U2Diff · Football-U | NLL: -2.16 -> -3.22 | Improved |
| compute Jacobian gradient during sampling (ablation) | U2Diff · Basketball-U | inference time (ms): 08 (x1.00) -> 31 (x3.88) | Hurt |
| use bi-variate instead of uni-variate Gaussian (ablation) | U2Diff with Jacobian · Basketball-U and NBA | AccRate: 86.3 -> 87.9 (Basketball-U); 80.9 -> 81.1 (NBA) | Mixed |
| use full Jacobian instead of diagonal approximation (ablation) | U2Diffine · Football-U | NLL: -3.22 -> -3.18 | No clear effect |
| bi-variate modeling over uni-variate in completion | U2Diff · Basketball-U | minADE20: 4.65 -> 4.59 | No clear effect |
| rank modes by AvgUcty instead of random | U2Diffine · NBA | Top-1 minSADE: 2.01 -> 1.93 | Improved |
| add RankNN error probability ranking over AvgUcty | U2Diffine · NBA | Top-1 minSADE: 1.93 -> 1.81 | Improved |
| add predicted variance as RankNN input (ablation) | RankNN · Soccer-U | mean Spearman rho: 0.55 -> 0.72 | Improved |
| add Multi-scene Transformer to RankNN (ablation) | RankNN · Football-U | mean Spearman rho: 0.47 -> 0.59 | Improved |
Closest papers by meaning
- Unified Uncertainty-Aware Diffusion for Multi-Agent Trajectory Modeling (CVPR 2025 · similarity 0.96)
- Sports-Traj: A Unified Trajectory Generation Model for Multi-Agent Movement in Sports (ICLR 2025 · similarity 0.88)
- Event2Tracking: Reconstructing Multi-Agent Soccer Trajectories Using Long-Term Multimodal Context (AAAI 2025 · similarity 0.87)
- SingularTrajectory: Universal Trajectory Predictor Using Diffusion Model (CVPR 2024 · similarity 0.85)
- Improving Transferability for Cross-Domain Trajectory Prediction via Neural Stochastic Differential Equation (AAAI 2024 · similarity 0.85)
- Partially Observed Trajectory Inference using Optimal Transport and a Dynamics Prior (ICLR 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.
