Spatio-temporal planning has emerged as a robust methodology for solving trajectory planning challenges in complex autonomous driving scenarios. By integrating both spatial and temporal variables, this approach facilitates the generation of highly accurate, human-like, and interpretable trajectory decisions. This paper presents a novel learned planning model-based spatio-temporal behavior decider, engineered to produce optimal and explainable driving trajectories with enhanced efficiency and passenger comfort. The proposed decider systematically evaluates the action space of the ego-vehicle, selecting the trajectory that optimizes overall driving performance. This method is particularly significant for autonomous driving systems, as it ensures the generation of human-like trajectories while maintaining high driving efficiency. The efficacy of the proposed framework has been comprehensively validated through rigorous simulations and real-world experimental trials on a commercial passenger vehicle platform, demonstrating its practical utility and performance advantages.
Huang et al. (Thu,) studied this question.