Abstract The paper proposes a novel approach toward EEG-driven position control of a robot arm by utilizing motor imagery, P300 and error-related potentials (ErRP) to align the robot arm with desired target position。 In the pro- posed scheme, the users generate motor imagery signals to control the motion of the robot arm。 The P300 waveforms are detected when the user intends to stop the motion of the robot on reaching the goal position。 The error potentials are employed as feedback response by the user。 On  detection of error the control system performs the necessary correc- tions on the robot arm。 Here, an AdaBoost-Support Vec- tor Machine (SVM) classifier is used to decode the 4-class motor imagery and an SVM is used to decode the presence of P300 and ErRP waveforms。 The average steady-state error, peak overshoot and settling time obtained for our pro- posed approach is 0。045, 2。8 % and 44 s, respectively, and the average rate of reaching the target is 95 %。 The results obtained for the proposed control scheme make it suitable for designs of prosthetics in rehabilitative applications。84906

Keywords Brain–computer interfacing · Motor imagery · P300 · Error-related potential · Position control of robot arm · Electroencephalography

S。 Bhattacharyya () · D。 N。 Tibarewala

School of Bioscience and Engineering, Jadavpur University,

Kolkata 700032, India

e-mail: saugatbhattacharyya@live。com

A。 Konar

Department of Electronics and Telecommunication Engineering, Jadavpur University, Kolkata 700032, India

1

Introduction

Non-invasive methods of recording brain signals are widely used in brain–computer interfacing (BCI) research for con- trol applications。 From the available noninvasive means of measuring brain signals, electroencephalography (EEG) is widely used among BCI researchers because of its higher temporal resolution, portability, availability and inexpen- siveness [21]。 BCI technologies aim at decoding brain sig- nals to detect the cognitive tasks executed by a user。 A few well-known brain signal modalities include steady-state visually evoked potential (SSVEP), slow cortical potential (SCP), P300, event-related desynchronization/synchroni- zation (ERD/ERS) and error-related potential (ErRP) [8, 22]。 The selection of brain modality is an important  issue in EEG-BCI analysis, and it depends on the cognitive task performed by the subject。 ERD/ERS originates during motor planning, imagination or execution (also referred to as motor imagery signals) [26], ErRP has shown promising results in detection of visually inspired errors [7], and P300 are detected during identification of rare (single) state from a given set of multiple states [11]。 Thus, these signals have relevance in our present study。

Motor imagery-based BCI are highly relevant for usage in rehabilitative applications which includes control of prosthetic devices [16] and controlling motion of wheel- chairs [18]。 Other applications include mind-driven motion control of mobile [3] and humanoid robots [6], thought- controlled navigation in virtual reality  environment  [2] and mind-controlled gaming [5]。 Its merit lies in the auto- matic control of external devices without neuromuscular intervention。

One of  the  open  areas  of  BCI  research  is  designing a control strategy for rehabilitative applications。 Early research on BCI employed a single signal modality, such as

P300, SSVEP and ERD/ERS for control applications [27, 28]。 But in recent studies, hybrid BCIs (i。e。, detection of at least two brain modalities in a simultaneous or sequen- tial pattern) have been emphasized for control applications because of its effectiveness over conventional BCI [18]。 For example, Pfurtscheller et al。 [25] used a motor imagery- based switch to turn ON/OFF an SSVEP-based BCI。 Long et al。 [18] have used motor imagery and P300 signals for continuous 2D cursor control and the direction and speed of a wheelchair, respectively。 Ferez et al。 [12] used ErRP signal to detect errors in motor imagery-based (open loop) cursor-position control。