Research map / Probabilistic and scientific modeling / Molecules and proteins
RAG-TP: A General Framework for Vehicle Trajectory Prediction via Retrieval-Augmented Generation
Ziyi Wang, Yang Zhang, Guijian Tang, Chao Zhang, Shibo Zhang, Xueqiong Li, Shaowu Yang
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 retrieval from a knowledge base (without map information) (ablation) | RAG-TP base model · Argoverse 2 validation | minADE6: 0.81 -> 0.73 | Improved |
| add map information to the RAG knowledge base (ablation) | RAG-TP · Argoverse 2 validation | minADE6: 0.73 -> 0.60 (full model) | Improved |
| use behavioral K-Means clustering instead of random sampling to build the knowledge base (ablation) | RAG-TP · Argoverse 2 validation | minADE6: 0.68 -> 0.60 | Improved |
| add Retrieval-Driven MoE fusion (ablation) | RAG-TP · Argoverse 2 validation | minADE6: 0.65 -> 0.60 | Improved |
| plug RAG module into MTR | MTR · Argoverse 2 | minADE6: 0.82 -> 0.76 | Improved |
| plug RAG module into Forecast-MAE | Forecast-MAE · Argoverse 2 | minADE6: 0.80 -> 0.73 | Improved |
| replace online map with retrieved priors (map-free RAG-TP vs MLB-Traj) | RAG-TP · Argoverse 1 map-free | minFDE6: minADE6 0.77 -> 0.72; minFDE6 1.25 -> 1.31 | Mixed |
| retrieve from the target domain's knowledge base (zero-shot, trained on Argoverse 2, evaluated on WOMD) (ablation) | RAG-TP · WOMD zero-shot | minADE6: 1.32 -> 1.05 | Improved |
| retrieve from the target domain's knowledge base (zero-shot, trained on WOMD, evaluated on Argoverse 2) (ablation) | RAG-TP · Argoverse 2 zero-shot | minADE6: 0.98 -> 0.75 | Improved |
| train and retrieve on unified Argoverse 2 + WOMD dataset (ablation) | RAG-TP · Argoverse 2 / WOMD cross-domain | minADE6: 0.67 -> 0.63 (AV2); 0.72 -> 0.69 (WOMD) | Improved |
| add RAG retrieval and fusion at inference | RAG-TP · Latency, RTX 4090 | End-to-end latency (ms): 23.6 -> 35.9 ms; memory 4.8 GB | Hurt |
Closest papers by meaning
- LaKD: Length-agnostic Knowledge Distillation for Trajectory Prediction with Any Length Observations (NeurIPS 2024 · similarity 0.87)
- DeMo: Decoupling Motion Forecasting into Directional Intentions and Dynamic States (NeurIPS 2024 · similarity 0.87)
- Towards Generalizable Trajectory Prediction using Dual-Level Representation Learning and Adaptive Prompting (CVPR 2025 · similarity 0.86)
- SmartPretrain: Model-Agnostic and Dataset-Agnostic Representation Learning for Motion Prediction (ICLR 2025 · similarity 0.86)
- C2F-TP: A Coarse-to-Fine Denoising Framework for Uncertainty-Aware Trajectory Prediction (AAAI 2025 · similarity 0.86)
- Leveraging SD Map to Augment HD Map-based Trajectory Prediction (CVPR 2025 · 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.
