Towards Safe and Reliable Autonomous Driving: Dynamic Occupancy Set Prediction

In the rapidly evolving field of autonomous driving, accurate trajectoryprediction is pivotal for vehicular safety. However, trajectory predictionsoften deviate from actual paths, particularly in complex and challengingenvironments, leading to significant errors. To address this issue, our studyintroduces a novel method for Dynamic Occupancy Set (DOS) prediction, enhancingtrajectory prediction capabilities. This method effectively combines advancedtrajectory prediction networks with a DOS prediction module, overcoming theshortcomings of existing models. It provides a comprehensive and adaptableframework for predicting the potential occupancy sets of traffic participants.The main contributions of this research include: 1) A novel DOS predictionmodel tailored for complex scenarios, augmenting traditional trajectoryprediction; 2) The development of unique DOS representations and evaluationmetrics; 3) Extensive validation through experiments, demonstrating enhancedperformance and adaptability. This research contributes to the advancement ofsafer and more efficient intelligent vehicle and transportation systems.