| Title |
Detailed-view Autonomous Vehicle Trajectory Prediction with Stratified Transformers |
| Authors |
응웬앙린(Ngan Linh Nguyen) ; 유명식(Myungsik Yoo) |
| DOI |
https://doi.org/10.5573/ieie.2024.61.4.3 |
| Keywords |
Autonomous driving; Trajectory prediction; Deep learning; Transformers; Spatial context encoding |
| Abstract |
Accurately predicting the trajectory of an autonomous vehicle is essential for ensuring safe navigation, but it can be challenging to predict multimodal future trajectory. Recent approaches, such as attention and transformers, have achieved state-of-the-art performance by considering agent interactions and map contexts. In this study, the focus was on trajectory prediction using an agent-centric approach with transformers, which provides a comprehensive understanding of the environment. Our proposed approach introduces a detailed-view spatial context encoding that arranges nearby agents or lanes into levels and encoding the information of each level, which enables the efficient encoding of complex spatial relationships. We adopt the transformer model to our proposed trajectory prediction scheme. Our approach was evaluated on the Argoverse benchmark and outperformed the state-of-the-art baseline in terms of accuracy. |