i。e。, move right, move left, move forward or no move- ment。 These outputs produce their corresponding com- mands necessary to drive the robot, which is shown in Table 1。 For example, if the user wants to move the robot

counterclockwise, he would imagine moving his hand kin- esthetically in the left direction。

The user observes  the  movement  of  the  robot  arm and  visually  detects  the  occurrence  of  error  during  the

Table 1 Mental tasks and their corresponding control commands Mental task Control commands

Left movement imagery Move counterclockwise Right movement imagery Move clockwise Forward movement imagery Move forward No Movement Pause at current position and wait for next command

Focus on target position (P300) Stop robot movement Detection of ErRP Perform necessary correction

movement of the robot。 If the robot moves in a different direction not intended by the subject (directional error), the robot would traverse back to its previous position and the subject would need to rethink the command again。 When the subject visually detects the link-end of the robot arm has reached or crossed the target, a P300 waveform is gen- erated which stops the movement of the robot arm。 If an ErRP is detected after the detection of P300 signal, it means that the link-end of the robot arm has crossed the desired position (positional error)。 In this case, the controller would attempt to realign the link-end by an offset (experimentally determined)。 The detection of the visual error is decoded from the EEG located at Fz, and the P300 signal is detected by the electrode at Pz。 It must be noted from the scheme that when the P300 or ErRP detector is activated, then no outputs from the MI classifier are accepted by the motor driver logic, as explained in Fig。 2b。

Experiments and data processing

This section describes the experiments undertaken to design the three detectors: MI, P300 and ErRP, shown in Fig。 2a。 Each subsection describes the stimuli designed for the offline and online sessions (as shown in Fig。 3), creation of the feature vectors [4] and the classifiers designed for each detector。 Five normal right-handed subjects in the age- group of 22–28 years have participated in this experiment。

2。3。1Design of the motor imagery detector

A visual stimulus is designed to train the user to perform the different MI tasks。 Here, the stimuli for offline sessions begin with a 30-s recording of baseline EEG followed by a repetition of trials, where the instructions on the MI tasks are given to the user。 The timing scheme of an inpidual trial, given in Fig。 3a, is as follows: a fixation cross for 1 s, followed by a MI task for 3 s and a blank for 2 s。 The fixa- tion cross alerts the user to get ready for subsequent motor imagination task。 The MI task here instructs the user to perform any one of the four tasks: move left, move right, move forward and no movement。 For training, the users performed the tasks in three separate sessions over every alternate days with 50 repetitions of each task in a random order。

The EEG signals acquired from the C3 and C4 locations are spatially filtered by Laplacian method [8] to remove the effect of neighboring electrode。 Here, we subtract    the

Fig。 3 The timing scheme of the visual stimuli。 a during training for motor imagery task, b during online testing for motor imagery task, c stimuli used in P300 task, d during online testing of P300 task, e during training for error detec- tion task, and f during online testing for error detection task。

(“+” indicates fixation cross, arrows represent the motor