Fig 1.
27 EEG electrodes recorded brain activities during a motor imagery task. Subjects watched first-person images of a robot’s hands through a head mounted display. A lighting ball in front of the robot’s hands gave motor imagery cue and subjects held images of a grasp for their own corresponding hand. Classifier detected two classes of results (right or left) and sent a motion command to the robot’s hand.
Fig 2.
Two different robots, a pair of very humanlike android hands versus a pair of robotic arms were used for BCI-teleoperation to evaluate the effect of humanlikeness and body ownership illusion on motor imagery learning.
Fig 3.
Subjects performed one session of non-feedback calibration (for the setup of subject-specific classifier) and one session of training with a normal computer screen. Then, they wore head-mounted display and prepared for BCI-teleoperation of the robots. Subjects were divided into two groups: (i) Geminoid group was first trained with a humanlike robot and then operated a pair of metallic gripper. (ii) ArmRobot group only used the metallic grippers for operation in both Sessions 3 and 4. All sessions included 40 trials. Subjects answered a questionnaire at the end of each teleopreational session.
Fig 4.
Median values and interquartile ranges for Q1 (self-evaluated performance) and Q2 (body ownership illusion) at the end of Session 3 and Session 4.
Fig 5.
Results for EEG offline analysis.
ΔJ (a measure of right/left class separation in the feature space during the motor imagery task) was calculated between two halves of Session 3 to evaluate the trend of motor imagery learning due to the BOT illusion. Also, ΔJ was compared between Session 3 and Session 4 to estimate the robustness of motor imagery skills learnt through each robot. Median values and interquartile ranges are depicted on each graph.