RUFI: Reducing Uncertainty in Behavior Prediction with Future Information


Autonomous driving has shown significant progress in recent years, but accurately predicting the movements of surrounding traffic agents remains a challenge for ensuring safety. Previous studies have focused on behavior prediction using large-scale data with diverse information like lane and agent information. However, these studies only use historical information, leading to uncertainty in predicting interactions between agents, which can result in collisions or incorrect trajectory predictions. To address this, we propose a novel method that uses future information during training to reduce uncertainty. Our approach leverages a Teacher-Student technique and attention-based model to reflect agent interaction. To bridge the gap in future information between the student and teacher models, we introduce Lane-guided Attention Module(LAM) that predicts trajectory using only local information in the student model. Our proposed model achieves state-of-the-art performance on the Argoverse motion forecasting dataset, demonstrating that future data, which was previously used only for supervision, can be effectively incorporated into the training process. This study is the first attempt to use future information during training for a behavioral prediction task, and provides a valuable contribution to this field.

In Conference on Computer Vision and Pattern Recognition Workshop 2023