Condition-Aware Trajectory Prediction via Velocity-Aware Model Selection
IEEE Access, cilt.14, ss.159882-159897, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 14
- Basım Tarihi: 2026
- Doi Numarası: 10.1109/access.2026.3744425
- Dergi Adı: IEEE Access
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Compendex, INSPEC, Directory of Open Access Journals
- Sayfa Sayıları: ss.159882-159897
- Anahtar Kelimeler: Autonomous driving, condition aware, trajectory prediction, velocity profile
- Hacettepe Üniversitesi Adresli: Evet
Özet
Safe and proactive decision-making in autonomous driving systems depends on predicting the future locations of vehicles, pedestrians, and other dynamic actors in traffic with high precision. However, trajectory prediction is an extremely challenging problem due to the existence of multimodal possible futures, the random nature of human behavior, and the complexity of densely interactive urban scenes. The previous trajectories of the actors are essential in future prediction. Usually, interactions between actors, map information, and environmental context are also considered. In this study, we focus on the velocity of the ego-vehicle for trajectory prediction as an additional factor. Our approach estimates future trajectories by selecting an appropriate velocity-aware model, which learns behavioral patterns from past experiences. The effectiveness of the proposed approach, the results of extensive experiments carried out on large-scale autonomous driving datasets covering different traffic scenarios and driving contexts, demonstrate that the proposed design reduces prediction error relative to the corresponding generalist baselines. The code and dataset splits generated for this study will be released on GitHub and can be accessed at https://github.com/mltmnrgz/velocity_aware_trajectory_prediction.