Abstract: CNC loading/unloading systems are often challenged by complex and dynamic environments, where camera-captured images contain excessive irrelevant features. This leads to numerous pseudo-edge points during edge detection, inaccurate workpiece positioning, and increased matching errors in grasping points, ultimately failing to meet high-precision machining requirements. To address these issues, this study designs a CNC loading/unloading system based on PLC-controlled intelligent collaborative robots. First, a compact high-performance PLC is selected as the core controller, and an online vision acquisition system hardware with an auto-switching visual probe is designed. Second, after preprocessing images captured by the collaborative robot’s vision system, sub-pixel edge localization technology is applied to partition and precisely locate pixel edges. Finally, leveraging the sub-pixel edge localization results, the PLC program controls the robot’s motion trajectory to achieve automated control of the entire loading/unloading process. Experimental results demonstrate that, during the grasping of bearing steel sleeves at five target points, the system achieves matching errors below 0.05 mm and positional errors consistently under 0.1 mm, fulfilling practical requirements. This approach significantly enhances the precision and reliability of robotic arm operations, demonstrating strong practical applicability.