Background: A brain-machine interface (BMI) should be able to help people with\ndisabilities by replacing their lost motor functions. To replace lost functions, robot\narms have been developed that are controlled by invasive neural signals. Although\ninvasive neural signals have a high spatial resolution, non-invasive neural signals are\nvaluable because they provide an interface without surgery. Thus, various researchers\nhave developed robot arms driven by non-invasive neural signals. However, robot arm\ncontrol based on the imagined trajectory of a human hand can be more intuitive for\npatients. In this study, therefore, an integrated robot arm-gripper system (IRAGS) that is\ndriven by three-dimensional (3D) hand trajectories predicted from non-invasive neural\nsignals was developed and verified.\nMethods: The IRAGS was developed by integrating a six-degree of freedom robot arm\nand adaptive robot gripper. The system was used to perform reaching and grasping\nmotions for verification. The non-invasive neural signals, magnetoencephalography\n(MEG) and electroencephalography (EEG), were obtained to control the system. The 3D\ntrajectories were predicted by multiple linear regressions. A target sphere was placed\nat the terminal point of the real trajectories, and the system was commanded to grasp\nthe target at the terminal point of the predicted trajectories.\nResults: The average correlation coefficient between the predicted and real trajectories\nin the MEG case was 0.705 �± 0.292 (p < 0.001). In the EEG case, it was 0.684 �± 0.309\n(p < 0.001). The success rates in grasping the target plastic sphere were 18.75 and\n7.50 % with MEG and EEG, respectively. The success rates of touching the target were\n52.50 and 58.75 % respectively.\nConclusions: A robot arm driven by 3D trajectories predicted from non-invasive\nneural signals was implemented, and reaching and grasping motions were performed.\nIn most cases, the robot closely approached the target, but the success rate was not\nvery high because the non-invasive neural signal is less accurate. However the success\nrate could be sufficiently improved for practical applications by using additional sensors.\nRobot arm control based on hand trajectories predicted from EEG would allow for\nportability, and the performance with EEG was comparable to that with MEG.
Loading....