The paper aims at designing a distinctive scheme of asynchronous position control of a robot arm by decod- ing motor imagery signals about four different states: left, right, forward and no movement。 On reaching the target (goal) position, the subject stops the movement of the robot arm by generating a P300 signal。 Ideally, the subject would reach the target with zero positional error。 But in practical scenario, errors would be generated during the control of movement by the robot which is detected by the presence of ErRP waveform in the EEG signal。 The novelty of the present research lies in the architectural design for posi- tional control of the robot arm。 The real-time performance of the proposed EEG-driven position control scheme has been studied using four metrics: percentage success rate, peak overshoot, steady-state (positional) error and settling time。

The rest of the paper is organized as follows。 Section 2 presents the methodology and experiments performed to design the EEG-driven position control scheme of a robotic arm。 Section 3 provides the experimental results of the pro- posed scheme during offline training paradigm, online test- ing paradigm and real-time robot arm control。 A summary of the proposed technique is discussed in Sect。 4 followed by the concluding remarks in Sect。 5。

2Methods and experiments

In this section, we discuss the proposed control scheme required to move a robot arm, the experiments leading to the scheme and, finally, a design description on the real- time controller。

2。1  EEG data acquisition

A NeuroWin (manufactured by NASAN) EEG machine with 19 electrodes (FP1, FP2, F8, F4, Fz, F3, F7, T4, C4, Cz, C3, T5, T6, P4, Pz, P3, T3, O2, O1) is used to under- take the experiments (Fig。 1)。 The EEG signals are ref- erenced to the right ear, and FPz  location  is grounded。 The  EEG  is  recorded  using  gold  plated  electrodes,  and

Fig。 1  10–20 electrode locations for the 19 selected channels

the impedances are kept below  5  kΩ。  The  EEG signals are amplified, sampled at 250 Hz, and band-pass filtered between 0。5 and 35 Hz。

It is well established from previous literature [8] that motor imagery (MI) signals originate from the cortical motor areas (primary motor cortex, sensorimotor area and pre-motor cortex), error-related potential originates from the cingulate cortex, and strong P300 signals are generated from the parietal region。 Based on the knowledge, we have processed the signals from C3 and C4 locations for MI detection, signals from Fz for error (ErRP) detection and signals from Pz for P300 detection。

2。2Design of control scheme

The basic block diagram to the proposed scheme is  given in Fig。 2a。 The proposed control scheme allows the move- ment (translation/rotation) of the robot arm in any random order as intended by the user。 Here, the control scheme employs three detectors for recognition of MI, P300 and ErRP brain states。 The MI detectors control the  direc- tional motion of the robot arm。 The P300 detector stops the motion of the robot arm when it reaches the goal posi- tion。 The ErRP detector detects the presence of directional (because of MI detector) or positional (because of P300 detector) error in the incoming signal, and on detection, the control system makes the necessary correction。 Here, the outputs of the different detectors are fed to the motor driver which controls the movement of the robot accord- ingly。 The logic followed by the motor driver is shown in Fig。 2b。

Here, the robot arm is capable of turning clockwise and counterclockwise, and translating in the forward direc- tion。 The user plans the movement of the robot arm and the acquired signals at C3, and C4 electrodes are decoded to  understand  the  motor  imagination  of  the  given  task, Fig。 2  a The block diagram of the proposed scheme。 b The instruction sequence of the motor logic driver