Research On Chinese Service Instructions Parser In Home Environment Based On Reinforcement Learning

With the continuous progress and development of our country's modern society,the aging of our country's population is gradually deepening,and the problem of home services for the elderly has become more and more serious.On the other hand,the increasing maturity of artificial intelligence technology has promoted the development of robots in an increasingly intelligent direction.Therefore,using robots to serve the elderly has become a feasible solution.In order for a robot to execute human instructions,it must first understand the content of the instructions.Therefore,it is necessary to study how to parse natural language instructions into machine instructions executable by the robot.The current research status at home and abroad was discussed,taking the family environment as the background,combining the relevant technologies in the field of natural language processing and deep reinforcement learning algorithms,the analysis of Chinese service instructions was discussed in thesis.Firstly,the service instructions commonly used in the home environment was collected,a corpus of Chinese service instructions commonly used in the home environment was constructed,and the necessary processing,including the word segmentation of the instructions and the generation of word vectors,as well as the necessary labels for the instructions was performed.After that,the instructions parsing problem was modeled and the action space,state space,and reward function was designed to enhance the application of learning.The problem of instructions parsing is regarded as a planning problem of the operation sequence in the instructions,and the problem of instructions parsing is turned into a problem of extracting operation names and operating parameters in the instructions in thesis.On this basis,we model and solve these two problems separately,and use the two models to extract the operation name and operation parameters in the instructions to form the operation sequence of the instructions.Secondly,a machine instruction library for mapping operation sequences to machine instructions was established in thesis.Firstly,the existing machine instructions in the Chinese service instruction library are added to the machine instruction library;secondly,the instructions that may appear but are not included in the machine instruction library can be updated by user to achieve the purpose of dynamically updating the machine instruction library.For the parsed operation sequence,the machine instruction planning system will match in the machine instruction library and map the operation sequence to the machine instructions.In addition,disambiguation processing is also required for ambiguous machine instructions.Finally,the feasibility of this idea was verified through experiments,and the operation sequences planner's use conditions and performance under different levels of complexity were analyzed.The machine instrucionts planner's planning ability was tested and verified by experiments analysis.