Journal Search Engine
The Journal of The Korea Institute of Intelligent Transport Systems Vol.24 No.6 pp.121-133
전방 차량의 선제적 기동 판단을 통한 신뢰 가능한 자율주행 시스템 연구
Anticipating Front-Side Vehicles’ Maneuvers for Trustworthy Autonomous Driving
Abstract
The advent of level 4+ autonomous driving exposes the limitations of threshold-based reactive planning and underscores the necessity of predictive strategies to ensure safety and ride comfort. This paper presents a framework that proactively forecasts the trajectory of the front-side vehicle to enable anticipatory decision-making. Inter-vehicle interactions in the front-side view were modeled as a spatiotemporal graph, and a Temporal Graph Cross Attention (TGCA) module performed cross-frame attention to fuse temporal trajectory cues while integrating edge weights to emphasize contextually influential front-side vehicles selectively, thereby refining the estimates of their trajectories and maneuver intent. Evaluated on real-world front-side dashcam footage, the approach consistently outperforms representative graph-based and sequence-based predictors across key metrics. These findings suggest that a spatiotemporal cross-attention design for TGCA effectively captures the interaction patterns in natural traffic scenes, supporting a transition from threshold-based control to anticipatory driving decisions informed by predicted front-side vehicle maneuvers.
