Design Of Multi Task Instruction Understanding System For Home Service Robot

With the combination of the Internet of things and artificial intelligence to form the artificial intelligence internet of things technology,as well as the arrival of China's aging society and the acceleration of the pace of life,the demand for home service robots also began to increase.The domestic and foreign home service robot industry market grows by 27% annually and is expected to reach 15.84 billion yuan in 2019.At present,25 of the 48 countries developing robots in the world are involved in the development of service robots,so the research focuses on home service robots Intelligent instruction understanding is a key area,and in view of the single task and security problems existing in the existing technology based on cloud computing for single task instruction understanding,a multi task instruction understanding system of home service robot at the edge is designed,which is characterized by using voice interaction,can understand the natural language instructions including multiple tasks,and deploy the reference model locally to protect users' privacy.The system adopts the C / S architecture,with user app and raspberry pie as a subsystem,client and server respectively,and raspberry pi and home device app as a subsystem,client and server respectively.The function of speech recognition is realized in user app,model reasoning is realized in raspberry pie,and communication between client and server is realized through Bluetooth.The research content of this paper is mainly reflected in the following aspects:Firstly,analyze the development status of artificial intelligence internet of things at home and abroad,the development of robot instruction understanding technology,natural language processing technology and multi label text classification algorithm at home and abroad is studied,and the multi task instruction understanding task is transformed into multi label text classification task.In the natural language processing technology,static word vector and pre training language model are studied.In the static word vector,statistical language model,One-Hot coding,neural network language model and word2 vec are studied.In the pre-training language model,ELMo,GPT,BERT and ALBERT are studied.Character is proposed as the input of the model and ALBERT is used as the pre-training language model of this experiment.At the same time,the deep neural network CNN,RNN,LSTM and Transformer are studied.According to the characteristics of CNN extracting sequence local semantic features by convolution calculation and Transformer extracting sequence global semantic features by Self-Attention calculation,so CNN-Transformer network is proposed to extract the semantic features of text sequences.Through the research of multi label text classification algorithm,a classifier based on soft attention mechanism is adopted.This paper studies the input and output characteristics of many kinds of activation functions,and proposes to use Re LU function in CNN layer and GELU function in Transformer layer.Finally,a CNN-Transformer attention model based on ALBERT is proposed.Secondly,a variety of multi task instruction understanding models are implemented,and the evaluation results of these models are analyzed.By comparison,the model proposed in this paper is the best,7.35% higher than CNN-LSTM model in accuracy matching rate,and 2.79% higher than CNN-Transformer model based on ALBERT.Third,the requirements of the system are analyzed,the overall structure of the system is designed,and the functions of the user mobile app,raspberry pie and home device app are completed.Then the system is simulated and the model operation is analyzed.Finally,the time required for the system to complete a process is about 700 ms,which meets the real-time requirements of the system.The integration of user app,raspberry pie and home appliance app constitutes a complete set of multi task instruction understanding system of home service robot based on edge position.Therefore,the research results of this paper have certain significance and reference value for promoting the intelligence of home service robot.