Answers are grounded in this paper and cite their sources.
Autonomous driving systems rely critically on accurate environmental perception, in which 3D object detection plays a fundamental role by enabling the recognition of object categories and their spatial positions through onboard sensors. With the rapid advancement of perception technologies, a wide range of 3D object detection methods have been developed, including recent approaches based on Transformer and Mamba architectures. However, a systematic and comprehensive survey of these emerging techniques remains lacking. In this paper, we present a comprehensive review of 3D object detection methods for autonomous driving across different sensing modalities, including camera-based, LiDAR-based, and multi-modal approaches. We analyze both the underlying theoretical principles and practical implementations, and systematically evaluate their strengths and limitations. Furthermore, we propose a novel taxonomy for multi-modal 3D object detection methods and provide an in-depth review of recent advances in Mamba-based architectures. Additionally, we discuss how these advanced methods improve detection accuracy and robustness in complex and dynamic driving scenarios. Finally, we summarize current research trends and highlight promising future directions to facilitate further developments in this field.
No takes yet. Share an insight, caveat, or question.
Shan et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: