Research On Motor Imagery EEG Recognition Technology In Robot Control

With the continuous development of robot technology,intelligent robots such as service robots,industrial robots and so on applied for various fields have entered the public’s vision.Especially mobile robots,such robots can help users to complete the remote grasp and delivery of items,which for the elderly and disabled people,mobile robots can become an important auxiliary tool in their daily life.At the same time,China has a large number of the elderly,as well as patients with physical disabilities and movement disorders.For such special groups,the traditional menu-based robot control mode can not effectively meet the higher service requirements and special application scenarios.The new robot control method based on motor EEG recognition technology can greatly improve this problem,and has become a hotspot of current research.In view of the above situation,according to the motion control requirements of mobile robots,the corresponding control module and instructions are designed from the user level.The key EEG processing in the control module was studied,and tow lightweight motor imagery EEG decoding models are proposed.The specific work is as follows: ⑴ Combined with the application scenarios of mobile robots,a control instruction system based on the combination of four-class motor imagery patterns,and the control experience of users was optimized for the instruction design problems existing the classic motor imagery brain-computer interface.⑵ Aiming at the shortcomings of traditional machine learning EEG classification algorithms and conventional convolutional neural network EEG decoding models,a lightweight channel-mixing convolutional neural network was proposed.The decoding accuracy can reach 74.9%.In addition,the feature visualization reveals its good interpretability and the generalization capability of decoding across the different datasets.⑶ Aiming at the serious resource consumption of deep,wide and fusion models in motor imagery EEG classification,a shallow double-branch convolutional neural network for EEG decoding was proposed to meet the real-time and miniatures of real brain-computer interface.Experimental results show that it achieves about 85% high accuracy classification on the open source four-class motor imagery EEG dataset with less resource consumption.In this research,the recognition technology of motor imagery EEG signals in robot control was studied,and the two proposed classification algorithms were evaluated on two public benchmark motor imagery EEG datasets.Experimental results show that the proposed algorithms obtained good decoding performance and can meet a specific mobile robot’s control instruction system,which provides a new idea for the motion control scheme of the robot.