An Mln-based Implied Intention Recognition With Selected Objects As Contextual Clues

The country is in the stage of accelerating population aging.Moreover,the large number of disabled persons increasing year by year has put forward a severe test to the nursing ability of family and society.To reduce the physical burden of disabled people and relieve the pressure of social care,this paper proposed a method to predict the implicit intention of action and automatically plan the household task according to the action action by locking the category information of household objects.The main contents include the modeling of object-action intention network based on Markov network,establishing the learning mechanism of user preference intention based on Q learning algorithm,and the research on the strategy of deconstruction and reorganization of household task operation skills based on the finite state machine.The specific content is as follows:Firstly,the self-care household tasks summarized in the International Classification of Functional Disabilities(ICF)were analyzed to extract the household objects and actions necessary for disabled persons to achieve self-care in the natural environment.Then,based on the concept of functional availability of objects under vision,the implicit action intention of objects is extracted to construct a data set.Then,Markov net algorithm is used to train the data set to construct the object-action intention net.Finally,the implicit action intention sequence of the locked object is solved based on the Viterbi algorithm to realize intention recognition.Secondly,in order to adapt to the household behavior habits of users,a user intention learning mechanism based on the Q learning algorithm is proposed to realize the automatic development of the intention network model in the online situation because the intention of the same object will change.Human-machine interaction is introduced into WMRA by constructing simple laser semantics,and then users express their decision on the output intention through interaction.Finally,the learning mechanism constructs a Q table based on the decision result and upgrades the intention network based on the Bellman equation so as to adapt to the user’s intention habit.Then,to realize the implementation of household tasks of the robot in the natural environment and solve the complex problems of actual motion modeling.Firstly,all the unit action libraries related to household tasks are established through remote bar teaching.Then,the trajectories of the deconstructed household tasks are generalized based on the DMP algorithm.Finally,the trajectories of the deconstructed household tasks are reconstructed based on the finite state machine,and the reconstructed trajectories are transmitted to the robotic arm to complete the household tasks.Finally,the above algorithms are integrated based on ROS and evaluated experimentally.Selection of ordinary household environment object,object-action intention recognition experiment to evaluate network on the ability to identify the quantity and category changes on learning performance experiment to assess learning mechanism in the face of different Numbers on the object the user habit adaptability,experimenting household tasks to evaluate the feasibility of WMRA automatic programming household tasks.