The article proposes a social-suitable and safety-sensitive trajectory planning (S4TP) framework for autonomous vehicles (AVs) that aims to achieve safe, efficient, and socially appropriate driving in complex traffic scenarios involving frequent interactions with human-driven vehicles (HDVs).
The key components of the S4TP framework are:
Social-Aware Trajectory Prediction (SATP): This module uses Transformer-based encoding and decoding to effectively model the driving scene and incorporate the AV's planned trajectory during the prediction process. It generates accurate predictions of the future trajectories of surrounding HDVs by considering their social interactions.
Social-Aware Driving Risk Field (SADRF): This module assesses the expected surrounding risk degrees during AV-HDV interactions, each with different social characteristics, and visualizes them as two-dimensional heat maps centered on the AV. It models the driving intentions of the surrounding HDVs and predicts their trajectories based on the representation of vehicular interactions.
The S4TP framework employs an optimization-based approach for motion planning, utilizing the predicted HDV trajectories as input. By integrating the SADRF, S4TP can execute real-time online optimization of the planned trajectory of the AV within low-risk regions, improving the safety and interpretability of the planned trajectory.
The proposed method has been validated through comprehensive tests in the SMARTS simulator, including challenging scenarios such as unprotected left-turn intersections, merging, cruising, and overtaking. The results demonstrate the superiority of S4TP in terms of safety, efficiency, and social suitability compared to benchmark methods.
На другой язык
из исходного контента
arxiv.org
Дополнительные вопросы