Sensor-Rich Robotic Grasping in Unstructured Scenes: A Review of Vision, Multimodal Fusion, Learning Architectures, and Closed-Loop Manipulation

Vision-based robotic grasping is a core capability for autonomous manipulation, yet reliable operation in unstructured scenes remains difficult because sensing, representation, contact mechanics, control, and hardware are tightly coupled. This structured qualitative review examines a 117-publication corpus spanning grasp detection and pose estimation, multimodal and tactile sensing, learning architectures, closed-loop correction, active interaction, and deployment-oriented evaluation. The update began with 91 candidate records; DOI and normalized-title deduplication produced 83 unique records, one tangential record was excluded, and 82 additions were combined with 35 prior publications. Primary-topic coding of the 82 additions identified grasp detection and pose as the largest group (26, 31.7%), followed by multimodal or tactile sensing (19, 23.2%) and learning or dexterous manipulation (11, 13.4%). Only three additions were primarily coded to datasets, benchmarks, simulation-to-real transfer, or closed-loop active perception, indicating a reporting and validation imbalance rather than the prevalence of individual failure modes. Across the corpus, geometry-centered perception remains effective when optical measurements and approach constraints are stable, whereas multimodal and active sensing become more important for transparent, reflective, deformable, occluded, or contact-uncertain objects. However, benchmark accuracy is not a proxy for robot-level reliability: latency, calibration drift, sensor dropout, collision clearance, grasp stability, recovery, and downstream task success can dominate physical outcomes. The review therefore links sensing failures to grasp representations, learning and fusion strategies, controller timing, evidence quality, and application constraints. Progress should be judged through context-preserving, fault-aware task protocols rather than isolated perception scores.