Research On Chinese Instruction Parsing Method For Home Service Robot

Human-robot interaction is of great significance to the development of home service robots.In order to achieve a more natural human-robot interaction,current research focuses on improving the accuracy and execution efficiency of robots in parsing natural language instructions.In recent years,the development of deep learning technology makes it possible to build a neural network model to solve the problem of instruction parsing,and good results have been achieved in the field of natural language instruction parsing.However,the research on Chinese instructions is not deep enough,with relatively few achievements.Therefore,we focuses on the related problems of Chinese instruction parsing in the home environment.Firstly,for multi-intention Chinese instruction parsing,we proposes a joint iterative multi-intention Chinese instruction action sequence extraction model in home environment,which aims to map multi-intention Chinese text instructions into robot action sequences.The model consists of two decoders and an encoder,among which two decoders are composed of the subject decoder and the object-intention decoder.The former analyzes the subject of the action sequence,and according to its parsing results,the latter analyzes the intention type and the object,and iteratively extracts from left to right in this process to improve the accuracy of multi-intention instruction parsing.Secondly,to solve the problem of key information loss in instruction sequence caused by insufficient data annotation,we constructs an end-to-end action sequence completion model based on attention mechanism.This model uses linear transformations with trainable weights to distinguish entities with different semantics to enhance the action sequence completion model.Attention mechanism and antagonistic training are added to improve the accuracy of prediction and robustness of the model.Finally,in order to further verify the proposed method,the process of the robot obtaining user instructions through Chinese voice and completing the tasks issued by the instructions is simulated.The simulation software and robot are selected according to the requirements,and the home environment is built with Gazebo.The speech recognition,speech synthesis,navigation and obstacle avoidance functions involved in the execution of the task are realized.Finally,the simulation demonstration is completed,which further proves the proposed method is feasible for practical application.