Trajectory generation on NumoSim (balanced split)
57.67BeSTAD AUROCGeo-CETRA
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Geo-CETRALearning=Rule-based decomposition, Backbone=Transformer2026.06 | 57.67 | 72.34 | 64.14 | 78.06 | 88.21 | |
| TrajGenAgentLearning=Zero-shot2026.06 | 50.08 | 52.55 | 53.68 | 56.9 | 63.98 | |
| Geo-LlamaLearning=Fine-tuning2026.06 | 33.75 | 70.46 | 60.57 | 78.88 | 89.62 |