Multi-Sensor SLAM and 3D Robotic Mapping in Confined Environments: A Systematic Review of Perception, Localization, and Autonomous Exploration

Low illumination, structural repetition, absence of satellite-based positioning signals, and communication constraints in confined spaces pose significant challenges for robotic localization, 3D mapping, and autonomous exploration. Following PRISMA 2020 reporting guidelines, this review included 166 studies identified through searches of Web of Science, Scopus, IEEE Xplore, and ScienceDirect. Bibliometric analyses—including publication and citation statistics, keyword co-occurrence, cluster classification, and temporal evolution—were conducted using VOSviewer, CiteSpace, and Bibliometrix. These quantitative approaches were complemented by content analysis to evaluate application scenarios, technical capabilities, and validation maturity. The findings indicate that existing research predominantly focuses on Simultaneous Localization and Mapping (SLAM), LiDAR perception, and multi-sensor fusion, with a strong emphasis on geometric mapping and localization. Conversely, investigations into semantic understanding, autonomous decision-making, risk assessment, and safe return mechanisms remain relatively insufficient. Validation methodologies are primarily limited to simulations and laboratory tests, resulting in scarce empirical evidence regarding real-world scenarios and long-term deployments. Based on these insights, this paper discusses technological evolution through the lens of scenario complexity, autonomy, and deployment maturity, and identifies prospective research directions such as semantic mapping, embodied intelligence, foundation models, multi-robot collaboration, risk-aware navigation, and standardized testing platforms. This review provides a reference framework for advancing robots in enclosed environments from geometric perception toward autonomous exploration.