Research On Motion Recognition By Force Myography Sensor Based On Bionic Robot

The FMG(force myography,FMG)control manipulator based on human-computer interaction,is a bionic manipulator device,using force myography to control.It is a fusion system of biosensor technology,pattern recognition technology and electronic information technology,with the trend of multi-pattern recognition,multi-sensor fusion,integration and portability.However,there is still a long way to go in terms of the application of FMG control in the aspects of control mode,sensor application,pattern recognition algorithm and control system.Therefore,in order to improve the generalization ability and accuracy of multi action pattern recognition of force myography sensor,this paper aims at improving the ability and accuracy of multi action pattern recognition based on force myography sensor,and hereby establishes a new type of interactive control method.The main research contents include: the development of the data acquisition system of FMG sensor and the embedded control system of the humanoid manipulator,the design and static performance calibration of the FMG sensor which based on the sandwich structure,the virtual simulation platform design of the humanoid robot,the gesture action pattern recognition algorithm,the influence of the change of the position and quantity of the sensor on the recognition accuracy,and the influence of the change of the grip force on the recognition system based on the FMG sensor.First of all,this paper makes a comprehensive review of the research status and development trend of the interactive control equipment at home and abroad.On this basis,a detailed analysis is conducted on the working principle,structure and performance of the human-computer interaction device based on the FMG sensor.Then,the overall framework of the FMG control manipulator system based on the pattern recognition method is established,including the sensor module,the data acquisition module,the embedded control system module,the pattern recognition algorithm module,the manipulator body and its virtual simulation module,and with the corresponding specific implementation methods.In this paper,the FMG sensor based on sandwich structure was prepared with QTC material as the sensitive material.The BP neural network algorithm(IMPGA)which was optimized based on density clustering genetic algorithm was designed and the task of nonlinear fitting to FMG sensor was completed.The experimental results show that the IMPGA algorithm which incorporates immigrant operators and dynamic domain search has the advantages of fast fitting speed,extensive generalization ability and good robustness.This paper proposed an incremental RBF neural network machine learning algorithm which based on K-means clustering algorithm was used to establish a humanoid robot pattern recognition control based on offline steady-state FMG signal data.The traditional RBF artificial neural network algorithm has slow training speed,through incremental algorithm based on impulse coefficient,and uses variable-scale grid search method to determine the learning rate and impulse coefficient of RBF artificial neural network algorithm,improve the training speed of RBF neural network algorithm.Based on this algorithm,we study the factors that affect the accuracy of motion recognition and propose a practical method to improve the recognition accuracy.The experimental results show that the RBF neural network incremental algorithm with impulse coefficient has the advantages of short training time,high recognition accuracy,and the network structure is simple,suitable for embedded systems.Based on the generalized inverse matrix method,the iterative formula of the output layer weights of the ELM algorithm is deduced and the online recognition experiments of the different gripper motions with multi-gesture are completed by using the extreme learning machine algorithm.And research the effects of the variation of the gripping force on the grasping motions recognition,the experimental results has some guiding significance for the construction of dynamic gesture recognition system.In this paper,a hardware experiment platform and a software virtual simulation platform are designed.The hardware platform adopts 2.4G wireless architecture,and a lightweight and low cost humanoid manipulator based on worm gear and worm drive with self-locking function is developed.The software virtual simulation platform based on LabView and Activex control combined with Matlab2012 realizes the on-line gesture recognition and control of the virtual humanoid robot.The experimental results show that the high rate of gesture recognition can be obtained by using the improved RBF neural network and wristband FMG sensor array.