Research map / Probabilistic and scientific modeling / Variational and generative sampling
Hierarchical Latent Structure for Multi-modal Vehicle Trajectory Forecasting
Dooseop Choi, KyoungWook Min
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 positional data preprocessing (speed, heading, tangent) (ablation) | baseline VAE forecaster · nuScenes | ADE_1/FDE_1: 3.15/7.53 -> 3.03/7.22 | Improved |
| add multi-lane vehicle-lane interaction (VLI) (ablation) | baseline + PDP · nuScenes | ADE_1/FDE_1: 3.03/7.22 -> 2.91/7.00 | Improved |
| add VLI (ablation) | baseline + PDP · nuScenes | ADE_15/FDE_15: 0.93/1.80 -> 0.94/1.82 | No clear effect |
| add vehicle-to-lane interaction (V2I) (ablation) | baseline + PDP + VLI · nuScenes | ADE_1/FDE_1: 2.91/7.00 -> 2.67/6.38 | Improved |
| add V2I (ablation) | baseline + PDP + VLI · nuScenes | ADE_15/FDE_15: 0.94/1.82 -> 0.91/1.77 | Improved |
| add GAN regularization (ablation) | PDP+VLI+V2I model · nuScenes | ADE_1/FDE_1: 2.67/6.38 -> 2.64/6.32 | Improved |
| add GAN regularization (ablation) | PDP+VLI+V2I model · nuScenes | ADE_15/FDE_15: 0.91/1.77 -> 0.89/1.72 | Improved |
| select neighbors within 5m instead of 1m of reference lane (ablation) | full model · nuScenes | ADE_15/FDE_15: 0.92/1.76 -> 0.89/1.72 | Improved |
| select neighbors within 5m instead of 10m of reference lane (ablation) | full model · nuScenes | ADE_15/FDE_15: 0.95/1.84 -> 0.89/1.72 | Improved |
| use distance-based selection (tau=5) instead of all vehicles (ablation) | full model · nuScenes | ADE_15/FDE_15: 1.00/1.98 -> 0.89/1.72 | Improved |
| use distance-based selection (tau=5) instead of most relevant vehicle only (ablation) | full model · nuScenes | ADE_1/FDE_1: 2.75/6.52 -> 2.64/6.32 | Improved |
| sample from all M modes instead of only the most prominent mode (ablation) | full model · nuScenes | ADE_15/FDE_15: 0.97/1.95 -> 0.89/1.72 | Improved |
Closest papers by meaning
- HPNet: Dynamic Trajectory Forecasting with Historical Prediction Attention (CVPR 2024 · similarity 0.86)
- ModeSeq: Taming Sparse Multimodal Motion Prediction with Sequential Mode Modeling (CVPR 2025 · similarity 0.85)
- C2F-TP: A Coarse-to-Fine Denoising Framework for Uncertainty-Aware Trajectory Prediction (AAAI 2025 · similarity 0.85)
- Learning Cooperative Trajectory Representations for Motion Forecasting (NeurIPS 2024 · similarity 0.85)
- CaDeT: a Causal Disentanglement Approach for Robust Trajectory Prediction in Autonomous Driving (CVPR 2024 · similarity 0.85)
- $K^2$VAE: A Koopman-Kalman Enhanced Variational AutoEncoder for Probabilistic Time Series Forecasting (ICML 2025 · similarity 0.84)
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.
