Research Of Humanoid Robot Control System Design And Implementation Based On Multimodal Brain Computer Interface

In recent years,the development of brain computer interfaces has provided a new way to communicate with the outside world for those people who suffer from severe disability.The brain-computer interface can avoid the human peripheral nervous system and muscle tissue,through directly communicate with the outside world by detecting the changes of human brain electrical activity.The traditional brain computer interface system mainly depends on a single type of electroencephalogram(EEG)signal as the input signal of the brain-computer interface,but the variety of control signals of this method can produce is limited which led to it only can do some simple recognition task.However the hybrid brain computer interface can greatly enrich the types of control signals through composed of different types of EEG signals,and has become an important direction of current brain-computer interface research.In order to solve the problem of less control signals of traditional BCI and could not apply in daily life,this thesis investigates a hybrid BCI system based on P300 and steady-state visual evoked potential(SSVEP)to control humanoid robot NAO to accomplish the task of moving as well as grabbing.The main research work of this paper is described as follows:(1)The design of experiment.This paper designs P300 and SSVEP experiment paradigm respectively.And the platform used in the experiment,the selection of the subjects,the selection of the electrodes,and various matters needing attention in the experiment were introduced in detail.(2)The preprocessing and feature analysis were performed for EEG signals under two paradigms respectively.First we used bandpass filter and artifact subtraction to remove noise and get the clean EEG signals from the collected EEG signals.In addition,the difficulties in identifying EEG signals and feasible solutions are pointed out by analyzing the amplitudefrequency characteristics of the SSVEP signal as well as P300 signal.(3)According to the characteristics of SSVEP signal and P300 signal,this paper came up with methods based on the traditional recognition algorithm to improve the recognition accuracy and solve the problem of difference between individuals.For the identification of SSVEP signals,this paper combines the canonical correlation analysis(CCA)with a convolutional neural network to design convolution CCA to identify the SSVEP signal,which makes the model have the ability of few-shot learning while improving the recognition accuracy.For the identification of P300 signals,due to the difference of P300 signals between people,this paper used transfer learning combined with support vector machine(SVM)to recognize EEG signals which improve the generalization ability of the model and P300 recognition accuracy.(4)In this paper,a novel hybrid BCI system is designed by using the previous EEG signal analysis method to control of the movement and grasping of the humanoid robot NAO.In the hybrid BCI system,all six participants participated in the verification experiment.During the experiment,all the subjects were able to use the EEG signal to manipulate the humanoid robot NAO to complete the task according to the mission requirements,which verified the feasibility of the system.In summary,the hybrid BCI system which proposed in this paper can control the humanoid robot to move and grasp robustly and stability,and it provides a new idea for the application of brain-computer interface in the future.