Trajectory Prediction on MMAUD M300 (test)
1.61Positioning Error (m)Multi-Modal Deep Fusion Framework
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Multi-Modal Deep Fusion FrameworkLoss Function=SmoothL1, Post-processing Strategy=Outlier detection, Threshold=22026.01 | 1.61 | 3.13 | |
| Bidirectional Cross-Attention MechanismMethod Architecture=Multi-modal Bidirectional Attention, Loss Function=Smooth L1, Post-processing Strategy=Outlier Detection + Smoothing2026.01 | 1.67 | 1.38 | |
| Multi-Modal Deep Fusion FrameworkLoss Function=SmoothL1, Post-processing Strategy=Smoothing, Window Size=52026.01 | 1.75 | 1.59 | |
| Multi-Modal Deep Fusion FrameworkLoss Function=SmoothL1, Post-processing Strategy=None2026.01 | 1.78 | 5.26 | |
| Baseline modelMethod Description=Single-modal LiDAR with Kalman Filter2026.01 | 2.79 | 1.73 | |
| Multi-Modal Deep Fusion FrameworkLoss Function=RMSE, Post-processing Strategy=None2026.01 | 3.2 | 7.36 |