Research map / Probabilistic and scientific modeling / Molecules and proteins
TrajTok: What makes for a good trajectory tokenizer in behavior generation?
Zhiyuan Zhang, Xiaosong Jia, Guanyu Chen, Qifeng Li, Zuxuan Wu, Yu-Gang Jiang, Junchi Yan
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 spatial-aware label smoothing (ablation) | TrajTok · WOMD val (20% train) | Realism Meta: 0.7597 -> 0.7702 | Improved |
| add spatial-aware label smoothing (ablation) | K-disks · WOMD val (20% train) | Realism Meta: 0.7443 -> 0.7584 | Improved |
| add spatial-aware label smoothing (ablation) | TrajTok · WOMD val (20% train) | minADE: 1.3797 -> 1.3428 | Improved |
| add flip symmetry to vocabulary (ablation) | TrajTok · WOMD val (20% train) | Realism Meta: 0.7670 -> 0.7702 | Improved |
| add symmetric vocabulary construction (ablation) | K-means · WOMD val (20% train) | Realism Meta: 0.7476 -> 0.7526 | Improved |
| add symmetric vocabulary construction (ablation) | K-disks · WOMD val (20% train) | Realism Meta: 0.7584 -> 0.7610 | Improved |
| replace K-disks tokenizer with TrajTok | SMART · WOMD val (20% train) | Realism Meta: 0.7584 -> 0.7702 | Improved |
| build vocabulary on nuScenes instead of Waymo logs: K-disks (ablation) | SMART · WOMD | Realism Meta: 0.7584 -> 0.7350 | Hurt |
| build vocabulary on nuScenes instead of Waymo logs: TrajTok (ablation) | SMART · WOMD | Realism Meta: 0.7702 -> 0.7641 | Hurt |
| reduce logged data for vocabulary from 10^7 to 10^5 samples: K-disks (ablation) | SMART · WOMD | Realism Meta: 0.7584 -> 0.7442 | Hurt |
| reduce logged data for vocabulary from 10^7 to 10^5 samples: TrajTok (ablation) | SMART · WOMD | Realism Meta: 0.7702 -> 0.7675 | Hurt |
| coarser grid resolution 0.1 m for all agent types (ablation) | TrajTok · WOMD val | Realism Meta: 0.7702 -> 0.7687 | Hurt |
Closest papers by meaning
- Trajeglish: Traffic Modeling as Next-Token Prediction (ICLR 2024 · similarity 0.90)
- DeMo: Decoupling Motion Forecasting into Directional Intentions and Dynamic States (NeurIPS 2024 · similarity 0.86)
- BAT: Behavior-Aware Human-Like Trajectory Prediction for Autonomous Driving (AAAI 2024 · similarity 0.86)
- SMART: Scalable Multi-agent Real-time Motion Generation via Next-token Prediction (NeurIPS 2024 · similarity 0.86)
- Language-Driven Interactive Traffic Trajectory Generation (NeurIPS 2024 · similarity 0.86)
- 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.
