Aiming at the industry pain points of low operation accuracy, poor environmental adaptability and difficult large-scale promotion of agricultural robots in unstructured farmland environments, this paper systematically reviews seven frontier key technologies of intelligent agricultural robots: three-dimensional environmental perception, 3D point cloud registration, target pose estimation, motion planning, multi-robot collaboration, compliant operation control, and visual servoing. The mathematical principles, algorithm evolution and engineering application cases of various technologies are analyzed in depth, and the performance indicators and applicable scenarios of different technical solutions are quantitatively compared. Combined with global industrial development data and domestic engineering practice, the future development trends of "cloud-edge-end" integration, perception-decision-execution closed-loop, and human-robot collaboration are predicted, providing reference for subsequent research and industrialization of agricultural robots.