Abstract
While knowledge about the working environment expands the autonomy of a telerobot, human support is necessary to complement its autonomy. Especially, human intervention at robot motion level is required to support the flexible task execution intuitively. Since the motion level intervention changes the real world directly without modifying the robot knowledge, it causes the inconsistency between them.
This paper presents a method for the world model management using the motion understanding system which recognizes the meaning of robot motions commanded by the motion level intervention.
The system is designed to have a hierarchical structure composed of symbolizers and a rule based motion interpreter. The symbolizer extracts events in the task execution from such signals as the robot hand pose, finger width, grasping force and external force, and sends them to the interpreter. The event driven production system realizes efficient interpretation and ensures the readability of the rules and the extensibility of the system.
The experiment proves that the motion understanding system can recognize a pick and place task even from the fluctuating operator's task execution using a master-slave manipulator. It also shows that the consistency of the world model can be maintained automatically.
