This study presents a robotic system that is demonstrated to complete a six-sided 3D puzzle. The main components of our system consists of a SCARA, a stereo camera, a conveyor belt, motors controlled by Arduino boards, and a computing station. In addition to the hardware composition, a set of specified markers, namely ArUco, is used to determine the target position where the puzzle cubes should be placed. The main procedure of our proposed algorithm for realizing a 3D puzzle game includes the following stages: initializing robotic system, localizing the target positions through ArUco markers, detecting any puzzle cube in an image capture, extracting the upper surface of the detected cube, defining the drawing point and pose, and picking and placing the puzzle. After all the cubes of puzzle are placed on their target positions, two single-axis slider are used to put the cubes closely. The main achievements of this study is making the robotic have vision and then combining deep learning and image processing.