Figure 1.
Schemata of a subject assisted by an exoskeleton within a possible rehabilitation scenario.
The exoskeleton is controlled online via signals directly recorded from the user and an eye-tracking system. The support is context driven and can be realized in the real world or a virtual scenario.
Figure 2.
Schemata of the conducted experiments.
Subjects were asked to move their right hand from a flat micro switch board to a buzzer, while looking at a green fixation cross presented on a PC-monitor. Between two consecutive movements a minimum resting time of s had to be maintained. Too early movements were reported to the subjects by changing the color of the fixation cross to red for
ms. After
valid movements one complete run was finished.
Table 1.
Classification results for all 4 conditions.
Figure 3.
Distribution of prediction times for EEG-based (blue) and EMG-based (red) movement prediction.
Time point zero corresponds to the physical movement onset, the red line at time ms indicates the range up to where predictions based on EMG were allowed, for EEG predictions up to
ms before physical movement onset were allowed, again marked with a red line.
Figure 4.
Distribution of prediction times for EEG-based movement prediction including the non movement range.
Time point zero corresponds to the physical movement onset, the red line at time ms indicates the range up to where predictions based on EEG were allowed, the range from
ms to
ms corresponds to the no movement class, hence all predicted windows in that range count as FPs.
Figure 5.
Prediction results in TP- and FP-rate for Condition A (EEG) (blue), Condition B (EMG) (green), Condition C (EEG “OR” EMG) (black) and Condition D (EEG “AND” EMG) (red).
The mean TP-rate and FP-rate for all subjects is shown; the bars indicate the standard deviation. Note for the x-axis two scales are used, since the FP-rates for Condition B (EMG) and Condition D (EEG “AND” EMG) are very small compared to the other two conditions. The two vertical dashes within the x-axis label highlight the scale change.
Figure 6.
Prediction results in balanced accuracy for Condition A (EEG), Condition B (EMG), Condition C (EEG “OR” EMG) and Condition D (EEG “AND” EMG).
The mean balanced accuracy for all subjects is shown; the bars indicate the standard deviation.