Self-Discovered Intention-aware Transformer for Multi-modal Vehicle Trajectory Prediction

2026-04-08

AI Summary

This work addresses the limitations of existing vehicle trajectory prediction methods that often rely on predefined graph structures or explicit intent labels, which constrain model flexibility. The authors propose a pure Transformer-based architecture featuring a dual-branch decoupled design to separately predict future trajectories and estimate neighbor-aware intent probabilities. A residual offset mechanism is introduced to effectively model multimodal and temporally ordered trajectory distributions. By eliminating dependence on graph neural networks, the approach integrates unsupervised intent discovery with interaction modeling, thereby decoupling spatial relationship reasoning from trajectory generation. Experimental results demonstrate that the proposed method significantly improves prediction accuracy and successfully learns semantically ordered multimodal trajectory distributions.

Abstract

Predicting vehicle trajectories plays an important role in autonomous driving and ITS applications. Although multiple deep learning algorithms are devised to predict vehicle trajectories, their reliant on specific graph structure (e.g., Graph Neural Network) or explicit intention labeling limit their flexibilities. In this study, we propose a pure Transformer-based network with multiple modals considering their neighboring vehicles. Two separate tracks are employed. One track focuses on predicting the trajectories while the other focuses on predicting the likelihood of each intention considering neighboring vehicles. Study finds that the two track design can increase the performance by separating spatial module from the trajectory generating module. Also, we find the the model can learn an ordered group of trajectories by predicting residual offsets among K trajectories.

Problem

• vehicle trajectory prediction
• multi-modal
• intention-aware
• autonomous driving
• Transformer

Innovation

• Intention-aware Transformer
• Multi-modal Trajectory Prediction
• Residual Offset Learning
• Dual-track Architecture
• Graph-free Modeling