Robotic grasp pose estimation in unstructured environments remains a critical challenge in the field of robotics. Existing methods, like GraspNet, while effective, often overlook pivotal aspects such as an object’s weight distribution and inherent frictional forces. This oversight can lead to unstable grasping, especially evident when handling delicate items. To address this, this paper introduces a novel approach: center-guided grasp pose estimation. By leveraging deep learning techniques, the proposed method predicts an object’s center even in scenarios where point clouds from a single viewpoint are incomplete. This methodology prioritizes the object’s center, promoting more stable and precise grasps. Preliminary results indicate a marked improvement in grasp quality, underscoring the potential for safer and more effective robotic interactions in complex environments.