Research map / Multi-agent systems and games / Human-AI and goal-conditioned
ECTraj: Enhanced Consistency Training for Multi-Agent Trajectory Prediction
Alen Mrdovic, Qingze Liu, Danrui Li, Mathew Schwartz, Kaidong Hu, Sejong Yoon, Mubbasir Kapadia, Vladimir Pavlovic
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 six-shot (best-of-K) training instead of one-shot (ablation) | ECTraj · Argoverse 2 motion forecasting | FDE6: 1.03 -> 0.96 | Improved |
| add six-shot (best-of-K) training instead of one-shot (ablation) | ECTraj · Argoverse 2 motion forecasting | ADE6: 0.53 -> 0.50 | Improved |
| add fusion of ground truth with teacher output (ablation) | ECTraj · Argoverse 2 motion forecasting | FDE6: 0.99 -> 0.96 | Improved |
| add fusion of ground truth with teacher output (ablation) | ECTraj · Argoverse 2 motion forecasting | MR6: 0.14 -> 0.13 | Improved |
| fuse only midpoint/endpoint instead of full ground truth replacement (ablation) | ECTraj · Argoverse 2 motion forecasting | FDE6: 1.10 -> 0.96 | Improved |
| fuse only midpoint/endpoint instead of full ground truth replacement (ablation) | ECTraj · Argoverse 2 motion forecasting | ADE6: 0.57 -> 0.50 | Improved |
| ECT timestep scheduling instead of ICT (ablation) | ECTraj · Argoverse 2 motion forecasting | FDE6: 1.02 -> 0.96 | Improved |
| ECT timestep scheduling instead of ICT (ablation) | ECTraj · Argoverse 2 motion forecasting | MR6: 0.15 -> 0.13 | Improved |
| best-of-K on diffusion with mode closest to data (ablation) | OptTrajDiff · Argoverse 2 motion forecasting | FDE6: 1.31 -> 1.17 | Improved |
| best-of-K on diffusion with mode closest to noise (ablation) | OptTrajDiff · Argoverse 2 motion forecasting | FDE6: 1.31 -> 1.42 | Hurt |
| replace diffusion denoiser with consistency model (1 step) | ECTraj vs OptTrajDiff · Argoverse 2 motion forecasting | GFLOPs: 1311 -> 132 | Improved |
| replace diffusion denoiser with consistency model (1 step) | ECTraj vs OptTrajDiff · Argoverse 2 single-scenario latency (ms) | denoising latency (ms): 29.71 -> 2.97 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- DONUT — DONUT: A Decoder-Only Model for Trajectory Prediction (ICCV 2025) · better on 1 of 1 shared result
- DeMo — DeMo: Decoupling Motion Forecasting into Directional Intentions and Dynamic States (NeurIPS 2024) · 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
- C2F-TP: A Coarse-to-Fine Denoising Framework for Uncertainty-Aware Trajectory Prediction (AAAI 2025 · similarity 0.89)
- Unified Uncertainty-Aware Diffusion for Multi-Agent Trajectory Modeling (CVPR 2025 · similarity 0.88)
- Improving Transferability for Cross-Domain Trajectory Prediction via Neural Stochastic Differential Equation (AAAI 2024 · similarity 0.87)
- STraj: Self-training for Bridging the Cross-Geography Gap in Trajectory Prediction (AAAI 2025 · similarity 0.87)
- Certified Human Trajectory Prediction (CVPR 2025 · similarity 0.86)
- Diffusion-ES: Gradient-free Planning with Diffusion for Autonomous and Instruction-guided Driving (CVPR 2024 · similarity 0.86)
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.
