Brain-robot Shared Control Based on Motor Imagery and Improved Bayes Filter

Brain-controlled robots are an innovative means of interacting and can also provide new solutions for disabled and stroke patients to communicate with the outside world. Since the poor real-time performance and poor accuracy of brain-computer interface (BCI) is not precise to control the robot directly, in order to avoid damage to the robot and humans in the process, this paper designs a brain-robot shared control system based on brain-computer interface. The motion direction of the robot controlled via four types of motor imagery (MI) signals. Feature extraction of MI signals is performed using common space pattern (CSP) combined with local characteristic-scale decomposition (LCD). The classification results are obtained with the appropriate features processed by the spectral regression discriminant analysis (SRDA) classifier. The Bayes filter algorithm is used to implement the robot shared control method, the belief of the robot's motion direction is calculated, and then the control ratio of the robot's autonomous motion and the BCI are assigned automatically. Considering that each control instruction given by BCI cost at least 1.5 seconds. To achieve better control effect at the interval between two instructions, the relationship with two steps of Bayes filter is redesigned, even if a new control data is not received, the robot will continuously update the measurement according to the previous control data, assign a new control ratio and execute the corresponding instruction, so that the robot can continuously adjust the movement intention and proportion during the instruction interval of BCI. The control effect was verified by online experiments. Using the improved Bayes filter algorithm, the success rate of the experiment is greatly improved, and the number of instructions used in single trial is reduced by 50%.