차량간 상호작용이 고려된 멀티-헤드 어텐션 구조 기반 주변 차량 주행 경로 예측 알고리즘

Abstract

Predicting the trajectory of surrounding vehicles is important in developing safe and socially compliant self-driving cars. For this reason, interaction-aware vehicle trajectory prediction with uncertainty is essential before making a robust decision-making system and trajectory planner. In this paper, we present a probabilistic vehicle trajectory prediction algorithm, which is scalable, interpretable and accurate. Using past trajectory and the properties of the surrounding vehicles, the proposed model generates the distribution of the future predicted trajectory. Our model consists of a simple encoder-decoder architecture based on multi-head attention. Like human drivers, the model can learn which vehicles to focus on for accurate prediction without requiring supervision. Inter-vehicle interaction learning improves the interpretability of the prediction network. We demonstrate our model's performance using a challenging, naturalistic trajectory dataset, showing clear improvement in terms of positional error on both longitudinal and lateral directions.