Research On Object Grasp Strategies For Home Service Robot Guided By Ontology Knowledge

The research on home robots that can realize the grasp and operation on objects has scientific significance and practical value,which improves the intelligent service capability of robots and speeds up its development.However,related works at present are still in the development stage,and challenges remain in realizing the robot application in home environment.Grasping objects is the premise for a robot to perform complex tasks,and the proper selection of grasp strategies determines the success of task execution directly.Object attributes,tasks,and home environments are all factors of grasp strategies in daily life.But due to the diversity and formal difference in them,it is difficult to take all factors into consideration,leading to the irrationality in strategies.To address this problem,researches on the generation of grasp strategies oriented to home robots are made as follows:1.A multi-attribute object knowledge representation method with functional parts as the center is proposed,to make robots available for functional parts and relevant knowledge including category attributes,visual attributes,physical attributes,affordance attributes,state attributes,location attributes.Firstly,six attributes(category attributes,visual attributes,physical attributes,affordance attributes,state attributes,location attributes)are proposed to describe objects,as a manner of enhancing the robot cognition on objects comprehensively and in depth.Then ontology representation on objects is made at the class level,and a multi-attribute object knowledge representation template centered on functional parts is constructed.After that,identified object instances and their attributes are automatically represented as ontology instances to complete the construction of the object ontology model.Thereafter,automatic reasoning on absent object attributes is realized by designing three kinds of rules conditioned on SWRL(Semantic Web Rule Language).Finally,the feasibility of the proposed method is testified on aspects of the transformation from data to knowledge,the cost time of checking and deducing object knowledge,and the number of deduced objects.In addition,the applicability of the object ontology model is verified in the task planning experiment.2.The grounding of the ontology model in Web Ontology Language Description Logic(OWL DL)is analyzed,to make home robots interact with home environment through the ontology model.According to the analysis,a grounding method of the object ontology model is proposed.Firstly,an association method that correlates classes,attributes,instances is proposed by introducing the solution of symbol grounding problem and the syntax and semantic of OWL DL.Secondly,a method for grounding the object functional parts based on the SWRL rules is proposed,conditioned on the study on the grounding method related to functional parts in the object ontology model.Finally,the method is verified through 3D model construction,Constrained Planar Cuts segmentation and SWRL rule correlation.3.A knowledge representation and reasoning method of semantic grasp strategy is proposed,to make robots acquire the knowledge such as grasp location,grasp type,approach direction,grasp force,opening width automatically.Firstly,according to the analysis of human grasp behavior,grasp location,grasp type,approach direction,grasp force,opening width,trajectory constrain are treated as the core elements of the semantic grasp strategies,to enhance the applicability of semantic grasp strategies.Secondly,the representation methods of task,environment and semantic grasp strategy are proposed to construct corresponding ontology models.Thirdly,a collaborative reasoning mechanism is accustomed to address the insufficient reasoning ability of a single reasoning machine,allowing robots to reason out semantic grasp strategies from the probabilistic ontology model,which satisfy the conditions of object attributes,tasks,and environment constrains.Finally,the method validity is testified on aspects of the correctness of grasp strategies and the cost time related to the deduced grasp strategies.4.A method for grasping planning and control of object functional parts with semantic grasp strategies is proposed to make grasp meet the constraints from object,task and environment and reduce the difficulty of grasping with five-finger manipulator,by integrating semantic grasp strategies and grasp actions.Firstly,a hierarchical architecture of the object grasp including knowledge representation and reasoning is designed.Secondly,a high-level grasp planning method conditioned on JSHOP2(Java Implementation of Simple Hierarchical Ordered Planner)is proposed,to integrate sub-actions into the semantic grasp strategies.Thereafter,a grasp action planning and controlling method based on grasp strategies is proposed to reduce the difficulty in controlling five-finger manipulator by designing the manipulator pose at the grasping start point according to the grasp location and approach direction and at the grasping point conditioned on the grasp location,approach direction and grasp type.Finally,a simulation operation scene with a Tiago robot is built to carry out experiments of different tasks and environments for testing the method feasibility.