Abstract: We report on the development of a deep-learnt grasping algorithm, Q-PointNet, which is capable of determining an adequate strategy for grasping a partially exposed object in a stacked pile. The grasping strategy includes the gripper’s posture and the finger mode, whether two fingers or three fingers. Because our predicted outputs are quaternion and mode, we also explain fully how to utilize the hybrid loss function to reach our goal within the limited training dataset. Moreover, according to the pose prediction, we developed an algorithm to estimate object width, and it is used to adjust the width from finger to finger of the gripper. In the end, the grasp experiments and their operation flows are presented, and the results show that our approach can grasp specific objects with a high accuracy in a stacked scenario.