Research On Task Analysis And Sequence Planning Of Home Service Robot Based On Knowledge Interaction

With the improvement of living standards,home service robots have increasingly entered the public’s vision.Currently,in home service tasks,robots’ motion execution is mostly based on preset instructions for simple tasks,lacking the ability to solve complex and unknown tasks,resulting in poor versatility and an inability to plan task execution steps.Facing the everchanging home environment and personalized user needs,how to understand the diverse task instructions spoken by users and generate action sequences for complex service tasks that are suitable for specific home environments is the key to the intelligentization of service robots.Therefore,this article focuses on three key points:intelligent command parsing for home service tasks,action sequence generation,and knowledge interaction optimization.The specific work of this article is as follows:(1)To address the problem of insufficient understanding of task instructions by home service robots when facing user-spoken service requests,a task type recognition and keyword extraction method based on gated mechanism and prior knowledge is proposed,which can improve the accuracy of instruction parsing.For the service task request instructions spoken by users,they are first converted into natural language text format through speech recognition and then transformed into word vector sequences using the FastText embedding method.Then,based on the BiGRU and attention mechanisms,a joint model for task type recognition and keyword extraction based on gate mechanism and prior knowledge is proposed,which can parse the task type and key words contained in the instruction,as well as match multiple objects and attributes based on the relationship judgment and subject-object alignment relationship matching model,obtaining the pairing results of objects and corresponding attributes in the instruction.Finally,a comparative experiment is conducted on the instruction parsing method to verify its intelligence and accuracy.(2)To address the problem that home service robots are difficult to plan a reasonable action sequence when performing complex service tasks,a service strategy-based action sequence generation method is proposed,which can reduce the complexity of tasks and effectively guide the planning of action sequences.Since it is difficult to directly plan the action sequence for complex or unknown tasks,a service step represented by text is proposed as an intermediate state,that is,the service strategy.A knowledge-enhanced service strategy generation model is proposed,which takes the task type and keywords parsed from the instruction as extended input and theme-constrained text for guiding and controlling the service strategy generation,making it more consistent with the logic of home service tasks.After optimization,each step of the service strategy is planned as a sub-task for action sequence planning,a family domain description based on PDDL is defined,and an autonomous generation method for problem descriptions for sub-tasks is proposed.A task planner is used to obtain the robot’s executable action sequence.Finally,a comparative experiment is conducted on the knowledge-enhanced service strategy generation method guided by keywords,and the service strategy generation result is used as the input for action sequence planning verification experiments to verify the completeness and reliability of the method in complex task planning.(3)To address the problem of poor environment adaptability of generated service policies due to the complex and changing home environment,a knowledge-based human-machine interaction question and answer optimization method is proposed,which can optimize the service strategy by complementing and disambiguating it with current environment information to improve reliability.Starting from the needs of personalized household,a home service knowledge base containing knowledge about home objects,attributes,and sequences is established,and a human-machine interaction question and answering module based on this knowledge base is built.Combined with visual detection,the non-matching parts of the service strategy and the environment are disambiguated through question and answer,optimizing the service strategy.Finally,an experiment is conducted on this optimization method,and the method’s effectiveness is verified in a home environment.Experiments show that the method in this paper can effectively parse out user instruction information and improve the ability to complete complex task planning.