The Key Technologies In Wearable Human-Computer Interaction System Based On Bioelectricity

With the development of computer technology and robotics,the new type of implicit human-computer interaction technology breaks the traditional method of program control.But it takes the initiative to comprehend the human interactivity,such as voice recognition,gesture recognition,robot automatic programming and so on.In the robot application field,the bioelectrical-based wearable human-computer interaction technology is one of the important fields to realize the implicit human-computer interaction,because of its advantages of natural and convenient interaction,simple and easy to use and predictive interaction,etc.As a new interactive frontier technology,it widely used in some advanced robot interactive system.The key technologies of bio-electricity-based wearable human-computer interaction system can be divided into two major aspects:discrete motion identification and continuous motion estimation.Based on the current research on these two aspects,some important method and designs is presented by experiments and further validation.Finally,combining the two key technologies,collaborative robot demonstration programming interactive system is studied by multi-modal interaction system technology,which research is beyond the comprehensive application of the two technologies.Focusing on discrete recognition,the algorithm of gesture recognition under non-specific human beings is been researched,including the comparison of feature extraction and feature selection schemes.artificial feature fusion under multiple kernel learning based on relevance vector machine(KA_MKRVM)and automatic feature processing by deep parallel convolutional neural network algorithm(PCNN)are proposed under the non-specific data set.Among them,PCNN can effectively adapt to non-specific person identification environment and avoid over-fitting by the tricks of data agument and activation function normalization.Both classification methods obtained the results above 2%higher recognition rate than the traditional SVM algorithm in the non-specific data set.The feasibility of the application is validated through the online mobile robot control experiment.In continuous motion estimation,we focus on the most commonly used single joint continuous motion estimation algorithm.The open-loop joint motion estimation model based on the Hill muscular dynamic model and the forward dynamic model of the human-machine system was established.Especially for its sensitivity to load disturbance,it is not easy to be stable.A parameter identification algorithm based on load compensation and a measurement equation based on inertial sensor information are proposed.A closed-loop estimation model based on extended Kalman filter is established.Based on the interactive exoskeleton system of lower extremity rehabilitation,parameter identification experiments and open-loop-closed-loop motion estimation were performed in off-line environment.The proposed closed-loop estimation algorithm and load disturbance compensation strategy can work better under different loads.It can keep stable for a long time under this closed-loop algrithm.In the integrated application experiment,the multi-modal interactive method of humanrobot is researched.By combining the proposed non-specific human gesture recognition algorithm and motion estimation algorithm,the system can recognize gestures and hand movements accurately and naturally.In the verification experiment,a demonstration programming experiment was carried out in the collaborative robot interaction system.Robot programming by demonstration and reproduction can be achieved by statistically coding and decoding of multi-modal information.Finally,a Q learning algorithm based on Gaussian process is proposed to enhance the learning efficiency of the end effector operation in the process of reproduction.