Abstract: High-precision, real-time and intelligent spatial information acquisition serves as a core foundation for surveying and mapping science to support digital twin cities, infrastructure operation and maintenance, disaster emergency response, natural resource investigation and deep space exploration. Conventional mobile mapping technologies have formed mature workflows in open and regular scenes, but as mapping targets expand from surface open spaces to complex environments such as underground spaces, industrial facilities, disaster sites and extreme planetary surfaces, which are generally characterized by global navigation satellite system (GNSS) signal limitation, severe structural occlusion, obvious geometric degradation and high operational risks, the traditional passive data acquisition mode relying on manual route planning can no longer meet the demand for efficient, safe and continuously updated spatial information acquisition. The advancement of mobile robot and multi-source sensing technologies provides a novel technical paradigm for complex space surveying and mapping. Heterogeneous robot platforms equipped with LiDAR, cameras, inertial measurement unit and other sensors can access inaccessible or high-risk areas for spatial data collection, while technologies such as simultaneous localization and mapping (SLAM), semantic segmentation and autonomous exploration enable robots to achieve real-time localization, 3D mapping and scene understanding in GNSS-denied environments. Nevertheless, existing research still has notable stage limitations: Most studies take SLAM as the core, focusing on pose estimation accuracy and local mapping performance, while paying insufficient attention to the integrity, uniformity, measurability and quality controllability of mapping deliverables. Autonomous exploration methods mostly target spatial coverage or navigation accessibility, failing to fully incorporate mapping requirements on point cloud density, observation angle, overlap rate and structural integrity. Semantic understanding results are mainly used as map annotations, and the feedback mechanism from semantic cognition to exploration decision-making has not been fully established. From the perspective of the surveying and mapping discipline, we systematically review the development status and challenges of robot autonomous mapping in complex spaces, define its task connotation and evaluation dimensions, and construct a closed-loop technical system of “high-quality 3D observation—spatial/semantic map construction—cognitive feedback decision-making”. We further summarize key research progress in three core aspects: (1) 3D observation quality control covering multi-sensor observation mode, trajectory motion modeling and online accuracy evaluation with degradation detection. (2) Multi-source 3D mapping and semantic cognition including task-oriented map representation, geometricsemantic element extraction and cognitive map construction. (3) The autonomous exploration mechanism collaboratively driven by map quality, semantic cognition and high-level task objectives. Finally, we analyze the technical bottlenecks in typical application scenarios such as infrastructure operation, underground resource exploration and emergency rescue, and prospects three major development trends: evolving from geometric mapping to semantic cognitive mapping, expanding from single-robot operation to multi-robot collaborative mapping, and transforming from offline one-time mapping to continuously updated long-term autonomous mapping.