A2G: Leveraging Intuitive Physics for Force-Efficient Robotic Grasping

In object manipulation, movements are inherently restricted by object geometry and dynamics. Humans use an intuitive understanding of physics while grasping objects, resulting in an efficient application of manipulation force. This involves a 'common sense' awareness of how objects behave in the physical world - a fundamental aspect of human perception. Drawing inspiration from intuitive physics, we introduce a novel task-oriented grasp framework, A2G, to select force-saving grasps for manipulation tasks and can be generalized with objects with the same physics manipulation logic. The framework trains a neural network that associates manipulation force trends with the point clouds of the target object to imitate human intuitive physics perception. Furthermore, we integrate a grasp pose filter to acquire improved grasp poses for manipulation tasks. Multiple real robot experiments demonstrated the proposed method can effectively reduce the manipulation force to improve the success rate by selecting the appropriate grasp and its generalizability to unseen objects.