Researcg On Intelligent Human-machine Interaction Technology Of Robot Arm Based On MEMS Sensor Network

As a kind of mechanical device similar to the human arm,the robotic arm can be used to replace humans to operate some dangerous places where humans are restricted.Its emergence provides a way for human society to interact with the surrounding environment in a long distance.In the interaction process,it can be safe and efficient,and it can even complete some tasks that humans cannot possibly complete.At the same time,multi-degree-of-freedom manipulators play an increasingly important role in our production activities,which has the characteristics of high precision,fast speed,flexible maneuverability,limited scene and so on.In the field of industrial production,the robot arm can be used to achieve the purpose of efficient production and reduce production costs,and can replace humans to complete complex repetitive mechanical labor.In recent years,due to the rapid development of virtual technology,the research of human-machine interaction technology has gradually become a hot spot,and the intelligent human-machine interaction of the robot arm plays an important role in it.In this paper,the robot arm motion recognition is selected as the entry point,the multidegree-of-freedom robot arm and the human arm are mapped,and different arm motion commands are designed for classification and recognition.At present,the traditional arm motion classification and recognition mainly based on computer vision and motion sensors.The classification and recognition methods mainly include template matching method,hidden Markov model,motion trajectory reconstruction and neural network method in the field of deep learning.With the rapid development of MEMS motion sensors,motion recognition based on sensors has also become a hot topic.Therefore,based on the advantages of MEMS motion sensor,such as high sensitivity,good dynamic performance and high adaptability to the environment,the data of the arm's command actions is collected in this paper,movement classification and recognition are realized by use the XGBoost integrated learning algorithm.First,this paper designs a set of arm motion acquisition system based on MEMS sensor network,which includes hardware and software.The hardware part includes mobile terminals(wearable devices)and data receiving terminals(personal computers).Mobile terminal is a wearable device which uses three sensors in series to measure the data of shoulder joint,elbow joint and wrist joint of arm joint.The software part refers to the design of the PC host computer interface,which mainly function is to facilitate data classification storage.Subsequently,eight common static arm movements and ten dynamic arm movements were defined,and data collection was completed by multiple people.Then,in order to ensure the data can be recognized by the XGBoost integrated learning algorithm,it is necessary to preprocess the arm motion data,which includes the use of three-axis accelerometer information to extract effective actions,the use of smoothing filtering algorithms to filter the data waveform,and the use of fast discrete Fourier transform(FFT)to discretely sample and align the data.At the same time,in order to improve the accuracy and robustness of the model,it is necessary to perform feature engineering processing on the collected motion data,which includes abnormal data processing based on N standard deviations,feature enhancement based on time and frequency domains,feature normalization based on Min-Max standardization,and selection of action data features based on PCA dimension reduction.Thus,a set of feature engineering processing methods suitable for MEMS data is summarized.Finally,based on the XGBoost integrated learning algorithm,the static and dynamic arm movement data are classified and recognized separately.Then,based on the performance indicators of different dimensions,the confusion matrix of the test data set is obtained,and the recognition rate of each action is analyzed and summarized.At the same time,action classification and recognition are performed on different input data again,and the influence of different input data on classification and recognition is analyzed and summarized.