Transferring the semantic constraints in human manipulation behaviors to robots

摘要

In this study, we aim to help robots manipulate objects with the guidance of semantic constraints (the grasp location, grasp type, approaching way, trajectory constraint, grasp force, and opening width) which are learnt from human manipulation behaviors. In order to transfer the complex and uncertain relationships between the attributes of object and task and semantic constraints in human behaviors to robots. We propose a representation method of human behaviors in machine-understandable semantics and a collaborative reasoning mechanism. With the suggested semantic constraints from human behaviors, the robot manipulation can be completed under “consciousness” and proper for both the object and task. The shareability of the object attributes, primitive actions and semantic constraints makes the proposed method be generalized to new objects and even new tasks. Additionally, we find that the object’s parts can be grasped sequentially, according to the object’s state and the action to be performed.