Research map / Large language models / Reasoning and chain of thought
Rethinking Training and Inference for Trajectory Forecasting: Linking Winner-Take-All back to GMMs
Qiyuan Wu, Katie Luo, Bharath Hariharan, Wei-Lun Chao, Mark Campbell
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 |
|---|---|---|---|
| replace greedy selection with test-time weighted merging | Wayformer · NuScenes Prediction (M=5) | minFDE5: 8.287 -> 3.050 | Improved |
| replace greedy selection with test-time weighted merging | MTR-e2e · NuScenes Prediction (M=5) | minFDE5: 5.369 -> 2.848 | Improved |
| replace greedy selection with test-time weighted merging | Wayformer · Waymo Open Motion (M=5) | minFDE5: 10.994 -> 4.968 | Improved |
| replace NMS with merging | MTR-e2e · NuScenes Prediction (M=5) | miss rate5: 0.521 -> 0.535 | Mixed |
| replace merging with direct training of 5 modes | Wayformer · NuScenes Prediction (M=5) | minFDE5: 3.050 -> 2.976 | Improved |
| replace merging with direct training | MTR-e2e · NuScenes Prediction (M=5) | minFDE5: 2.848 -> 2.938 | Hurt |
| add one-step EM finetuning of mode probabilities before greedy selection | Wayformer · NuScenes Prediction (M=5) | minFDE5: 8.287 -> 5.702 (-31%) | Improved |
| add one-step EM finetuning before NMS | MTR-e2e · NuScenes Prediction (M=5) | minFDE5: 4.440 -> 3.752 (-15%) | Improved |
| add one-step EM finetuning before greedy selection | Wayformer · Waymo Open Motion (M=5) | minFDE5: 10.994 -> 10.370 (-5.7%) | Improved |
| add one-step EM finetuning before greedy selection | Wayformer · Waymo Open Motion (M=5) | miss rate5: 0.695 -> 0.757 | Hurt |
| add one-step EM finetuning | Wayformer · NuScenes Prediction (M=5) | top-5 NLL: 4404.5 -> 1796.4 (-59%) | Improved |
| add one-step EM finetuning | MTR-e2e · NuScenes Prediction (M=5) | top-1 NLL: 792.9 -> 777.0 (-2.0%) | No clear effect |
Closest papers by meaning
- Towards Metric-Agnostic Trajectory Forecasting (ECCV 2026 · similarity 0.89)
- DeMo: Decoupling Motion Forecasting into Directional Intentions and Dynamic States (NeurIPS 2024 · similarity 0.87)
- DriveGPT: Scaling Autoregressive Behavior Models for Driving (ICML 2025 · similarity 0.87)
- DriveFuture: Future-Aware Latent World Models for Autonomous Driving (NeurIPS 2026 · similarity 0.87)
- Motion Forecasting in Continuous Driving (NeurIPS 2024 · similarity 0.87)
- LaKD: Length-agnostic Knowledge Distillation for Trajectory Prediction with Any Length Observations (NeurIPS 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.
