The operating principle of the cursor task

[arvay]

Hi,all,

I'm trying to use the Cursor Task example for left and right control only.
I did the three images tasks(rest,Left hand,Right hand) and I've completed the offline analysis.
C4 was selected for left movement at 10Hz and C3 for right movement at 10Hz based on the r^2 values.
However, I don't understand the principle of operation of the cursor task and the configuration of the weights of the classifier,
so i don't get good results from the Cursor Task sessions.

Please help.
Thanks for advance,
arvay

[pbrunner]

Arvay,

the post below lists the specific steps to successfully implement mu/beta based cursor control. To get started you can use the parameter file that is listed along with the instructions under http://tinyurl.com/q83axf3

http://www.bci2000.org/phpbb/viewtopic. ... sket#p8146

Regarding your question for the classifier, I can tell you that the only thing that matters for a one-dimensional control is the sign of the classifier weight. In other words wether the power augmentation during the relaxation condition results in the cursor going up or down. To get you started, I would suggest that you begin with a screening that includes overt and covert movements (as detailed in the linked PDF). You then train the users skill by providing feedback in the basket task. This task has the advantage that it gives the user as much time as necessary to move the cursor in the correct target. In this training you also determine the mean and std of the signal using the Normalizer. After you feel that the user has sufficient control you can disable the adaptation and just crank up the gain factor to boost the speed of the cursor.

If you are moving towards two- or three-dimensional control you will most likely need a more complex classifier. For example, you can use the difference between mu and beta as the control signal for a third control signal.

Regards, Peter