Research And Implementation Of Mobile Robot Task Understanding Based On Natural Language

The mobile robot studied in this paper is a kind of service robot,and task understanding is a key technology for service robots.Task understanding is the process of converting tasks expressed by the user's natural language into atomic actions that the robot can perform,which includes two main processes,the first is human-robot interaction,and the second is task planning.This paper focuses on the human-robot interaction process.The difficulty of this process is mainly manifested in language understanding.The first problem is semantic parsing.Current methods to solve the problem either rely on a manually compiled rule base,which does not have the ability to learn and predict,or rely on an amount of training data,for which obtaining hundreds of thousands of human-robot interaction data is very difficult.This paper designs a parsing algorithm based on sentence similarity,which can effectively solve the problem of lack of data.Compared with support vector machines and recurrent neural network models,the performance of the algorithm on FBM3 dataset is significantly improved.Another difficulty in language understanding is that the semantic representation obtained through semantic parsing must be mapped into specific objects in the robot's working environment so that the robot can perform the task.This process is called the grounding process.In order to solve the grounding problem between semantic ontology and specific objects,this paper improves the grounding function by using knowledge graph,which obtained a good classification effect.The human-computer interaction of robots also needs to solve the problem of dialogue management,which is in fact a decision process.In this paper,a decision method based on Bayesian network is designed based on previous studies,which can effectively solve robot's decision problem under multiple impact factors.Finally,for the mobile robot in this paper working in a specific field and lacking open knowledge of task decomposition,the method of handwritten task semantic knowledge is used to solve the task planning problem.The main contributions and innovations of this paper include: designed a sentence similarity calculation method and semantic parsing method based on dependency grammar,which solves the problem of semantic parsing on robot's small sample dataset;using Bayesian network for probabilistic reasoning,solved the decision problem in multiple impact factor environment;improved an grounding function by combining knowledge graph,which solved the grounding problem between semantic ontology and specific objects.