Robotic object manipulation has seen remarkable advancements in recent years, with the potential to revolutionize industries ranging from manufacturing to healthcare. A critical aspect of successful manipulation is the generation of optimal grasp points, which directly influence the efficiency and reliability of robotic operations. This thesis delves into the domain of grasp planning and introduces a novel approach to grasp point generation leveraging the PointNet++ architecture. The proposed methodology harnesses the power of deep learning and point cloud data processing to identify and select grasp points that maximize the probability of success. Through an extensive review of the existing literature, we identify the limitations of traditional grasp planning techniques and highlight the advantages of utilizing deep learning models and point cloud data processing for this purpose.