Summary and Reflections on Pedestrian Trajectory Prediction in the Field of Autonomous Driving

Pedestrian trajectory prediction is a classic and challenging scientific task that involves complex engineering science and human factors. These challenges have spurred a significant amount of research, opening up new possibilities for practical applications such as autonomous driving. This importance stems from its pivotal role in ensuring safe and harmonious interactions between autonomous vehicles and pedestrians. Predicting pedestrian trajectories is an inherently challenging task, primarily due to the necessity for algorithms to capture and model the intricate interplay between pedestrians and their dynamic environmental surroundings. Furthermore, these algorithms must generate multi-trajectory distributions that effectively encapsulate the inherent uncertainty associated with pedestrian behavior. From an engineering application perspective, this paper reviews existing work along the input-method-output continuum to gain a deeper understanding of pedestrian trajectory prediction algorithms. The review encompasses a wide array of aspects, including the classification of interaction types, the examination of various prediction models, and the elucidation of diverse output types. Moreover, the paper undertakes the categorization of existing datasets based on distinct application scenarios and sorts out detailed evaluation metrics for trajectory prediction algorithms. Lastly, this study emphasizes the prevailing research challenges within the domain of pedestrian trajectory prediction, particularly in the context of autonomous driving. It also offers well-considered reference solutions to these challenges. The overarching goal is to underscore the vital role played by pedestrian trajectory prediction in the realm of autonomous driving and to catalyze further progress and innovation in autonomous driving technology through the provision of this summary and insightful analysis.