A knowledge-based approach for object classification for robotic assembly

This paper presents a system for object classification for robotic assembly. An integration of image data, grasping models and assembly analysis through a knowledge-based approach is described. The shape of an object is regarded as feature for object recognition based on image data. Grasping models are built from object features for classification of objects and for analysis of grasping features and behaviour. Based on the grasping models and assembly tasks, the grasping location and approach direction of a gripper is analysed and the assembly operation is properly carried out by the robot hand. Information transformation among image data, grasping models and assembly analysis is made through the knowledge-based approach. The proposed approach includes an object grasping knowledge representation and an assembly oriented reasoning mechanism. The knowledge representation consists of the object geometry, grasping feature and assembly task representation. The reasoning mechanism performs grasping and assembly analyses. It takes three steps: selection of candidates of grasping models from the system in terms of the object features, analysis of grasping features of the object to be grasped and analysis of assembly performance in terms of the assembly tasks and the grasping models. The knowledge representation and reasoning mechanism is organised as a hierarchical structure