<p>Spatio-temporal data plays an important role in intelligent traffic management and public service development. With the proliferation of location-aware devices, the collection of trajectory data has become increasingly convenient. However, the direct use of real-world trajectory data in downstream applications raises significant privacy concerns. To address this issue, trajectory generation has emerged as a promising solution by generating synthetic trajectory data. While existing surveys often focus narrowly on specific aspects or lack in-depth technical analysis, this paper presents a comprehensive review of the current research in this area. We categorize existing methods into two broad groups: model-based and learning-based approaches. Within each category, methods are further subdivided according to their underlying principles and evaluated based on their respective strengths and limitations. In addition, we summarize recent advances and outline potential directions for future research in this area.</p>