Trajectory Generation on NumoSim
0.0006DistanceTrajGenAgent
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| TrajGenAgentBackbone=Qwen2.5-32B-Instruct, Generation mode=zero-shot, Temperature=0.902026.06 | 0.0006 | 0.0993 | 0.0155 | 0.2117 | 0.0002 | 0.0002 | 0.0075 | |
| Geo-LlamaBackbone=Llama-2-7b-chat-hf, Fine-tuning=LoRA, Batch size=48, Learning rate=0.00001, LoRA r=16, Epochs=20, Temperature=1.22026.06 | 0.0075 | 0.2361 | 0.0028 | 0.0128 | 0.0001 | 0.0001 | 0.0087 | |
| TransformerTraining epochs=200, Optimizer=Adam, Learning rate=0.001, Embedding size=256, Architecture=decoder-only, Layers=4, Attention heads=42026.06 | 0.0082 | 0.2945 | 0.215 | 0.162 | 0.0079 | 0.0112 | 0.0118 | |
| SeqGANEmbedding dimension=16, Hidden units=16, Pre-training epochs=40, Adversarial epochs=202026.06 | 0.0085 | 0.0998 | 0.241 | 0.1585 | 0.0082 | 0.0107 | 0.012 | |
| Geo-CETRABeam size=10, Top k=3, Optimizer=Adam, Learning rate=0.012026.06 | 0.0093 | 0.3337 | 0.006 | 0.1128 | 0.0002 | 0.0002 | 0.0088 | |
| GRUTraining epochs=200, Optimizer=Adam, Learning rate=0.001, Embedding size=256, Layers=6, Hidden units=5122026.06 | 0.0111 | 0.2557 | 0.2145 | 0.1561 | 0.0137 | 0.0159 | 0.0156 | |
| LSTMTraining epochs=200, Optimizer=Adam, Learning rate=0.001, Embedding size=256, Layers=6, Hidden units=5122026.06 | 0.0146 | 0.3113 | 0.3013 | 0.1981 | 0.0893 | 0.0134 | 0.015 |