Human Trajectory Prediction (HTP) is a challenging task in computer vision and robotics, with numerous applications, including Autonomous Driving, Smart City Surveillance, Human-Machine Interaction, and Autonomous Robots. Despite significant progress in recent years, existing methods have focused on accuracy, social interaction modeling, and deterministic diversity. In contrast, little attention has been paid to uncertainty modeling, calibration, and forecasts from short observation periods, which are crucial for downstream tasks such as path planning and collision avoidance. Moreover, existing methods often rely on datasets and evaluation metrics that may not align well with the prediction goals of real-world applications, such as integrating HTP into autonomous vehicles or robot path-planning tasks. This survey provides a comprehensive review of HTP methods, datasets, and evaluation metrics, and discusses the need for more application-specific evaluation and the use of diverse datasets in relation to established rules and conditions. By providing a thorough understanding of the current State of the Art in HTP and its use cases, this survey aims to facilitate the development of more accurate and usable predictive models for real-world applications.
Hetzel et al. (Wed,) studied this question.