GACNet: Interactive Prediction of Surrounding Vehicles Behavior under High Collision Risk
Trajectory prediction technology is essential for driving safety in autonomous vehicles and is advancing rapidly. Current research mainly aims to enhance prediction accuracy in typical traffic conditions. However, less attention has been paid to low‐probability, high‐risk safety‐critical events. It...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Wiley
2025-05-01
|
| Series: | Advanced Intelligent Systems |
| Subjects: | |
| Online Access: | https://doi.org/10.1002/aisy.202401040 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Trajectory prediction technology is essential for driving safety in autonomous vehicles and is advancing rapidly. Current research mainly aims to enhance prediction accuracy in typical traffic conditions. However, less attention has been paid to low‐probability, high‐risk safety‐critical events. It is undeniable that multiagent interaction behaviors in safety‐critical events are more complex and difficult to predict. To address this, an interaction mechanism based on generative adversarial networks training, named GACNet, aimed at effectively predicting multiagent interaction behaviors under potential collision risks is proposed. GACNet is a deep learning framework capable of learning and capturing complex interaction patterns between multiple agents from real vehicle trajectory data. In addition, A conflict analysis module, which analyzes the predicted future trajectories and assesses potential collisions to provide a more detailed characterization of the interaction behaviors in safety‐critical events is designed and incorporated. This design enables the model to predict vehicle trajectory behaviors in safety‐critical events more accurately, aligning them more closely with real‐world trajectory distributions. This mechanism is validated in various highly interactive roundabout and urban road scenarios. The results demonstrate that GACNet accurately learns vehicle behavior characteristics from real traffic data and makes precise predictions of multiagent interaction behaviors in safety‐critical events. |
|---|---|
| ISSN: | 2640-4567 |