Effective human-robot collaboration requires social robots to adapt to individual human grasping habits to ensure smooth and safe object handovers. However, current robotic systems struggle to interpret diverse grasping behaviors, as individual habits can introduce variations even within the same grasp topology. This limitation affects the effectiveness of robotic systems in social contexts. This paper presents a grasp adaptation algorithm that enables robots to recognize and adjust to human grasping habits. The system identifies human grasping poses from RGB images and maps them to abstract representations consisting of 21 3D points each. These representations are then classified into one of six standard grasp topologies. Based on the identified topology, key points are selected from the abstract grasp to estimate the object’s pose. A reinforcement learning model is subsequently employed to optimize the object handover process. Experimental results demonstrate that this approach significantly enhances both the fluidity and safety of human-robot object handovers.