Research On Environment Adaptive Service Strategy For Home Service Robot

With the development of society and the advancement of technology,home service robots are becoming more involved in people’s lives increasingly.In recent years,users’ requirements for quality of life have gradually improved,resulting in more complex service requirements and more diverse environmental layouts.In the face of complex service requirements and diverse home environments,the key to improve service quality for home robots is how to obtain the service strategy,which are detailed,specific,environment-suitably task process-related text descriptions via the users’ natural language request.Therefore,this thesis explores the environment-adaptive service strategy for home service robots,and proposes a task analysis→strategy generation→optimization verification framework,for improving the home service robots’ ability to perform complex tasks in different home environments,and then enhance the success rate of task execution and task completion quality.This thesis mainly completes the following work:(1)Aiming at the problem that home service robots insufficient understanding of real task requirements when facing users’ service requests,a service task analysis method based on prior knowledge is proposed,which can increase the robots task recognition accuracy effectively.Firstly,a prior knowledge database is built by integrated relevant service text information using natural language processing tools combined with regular expressions,including task instruction database built for task analysis,keyword sequence database built for keyword sequence expansion and service strategy database built for model training;Then,a user speech recognition model based on the improved residual network unit and the bidirectional long shortterm memory network is built,which is used for collecting the users’ requests voice accurately,also for converting the voice information into text information in real-time;Finally,the mapping method of service request instruction,text to task,is proposed,including phrase entries are extracted by Stanford Parser to remove the interference of irrelevant information,and the weighted semantic representation based on Word2Vec vector which used to carry out task mapping combined with task instruction database to complete service task analysis.(2)Aiming at the problem of insufficient ability of home service robots to perform complex tasks in different environments,an adaptive service strategy generation model based on environmental information is proposed,which can adjust the content of the strategy by the current environmental information,and then which leads the robots to complete the task.Firstly,a keyword sequence expansion method based on the dynamic fusion feature vector is proposed,which uses bidirectional long short-term memory network fused multi-feature to obtain deep semantic representation,and combines keyword sequence database to supplement task-related information.Then,the family environment information matching process based on the ontology knowledge database is designed to work out the service keyword sequence that fits the current environment,which manages relevant environmental knowledge by Protege,which uses query mechanism to match environmental information and uses reasoning mechanism to replace environmental information.Finally,an adaptive service strategy generation method based on service keyword sequence guidance is proposed,which integrates topic coding as the input sequence representation,fine-tunes the strategy generation model by the service strategy database,and uses the Top-k sampling method to optimize the model decoding logic,so as to generate the service strategy with reference to the keyword sequence.(3)Aiming at the problem of task execution deviation caused by the interference of irrelevant information in the initial strategy,a service strategy optimization and verification mechanism based on task information is proposed,which improves the feasibility and reliability of the strategy.Firstly,a service strategy optimization network based on task instructions is constructed,which predicts strategy effective segment boundaries by a pointer network based on bidirectional attention flow to eliminate irrelevant information,and the boundary features with semantic features is integrated by the generative model based on bidirectional long shortterm memory network for generating optimized strategy;Secondly,a service strategy verification framework based on task topic is designed,which uses Transformer-XL to extract long-distance dependencies to generate strategy topics,and performs correlation calculation based on smooth inverse frequency word vectors combined with task instructions to complete service strategy verification.