Available online at www.sciencedirect.com
ScienceDirect Procedia Computer Science 84 (2016) 147 – 151
7th International conference on Intelligent Human Computer Interaction, IHCI 2015
Trajectory path planning of EEG controlled robotic arm using GA Rinku Roya, M. Mahadevappab*, C.S.Kumarc a
Advanced Technology and development centre, Indian Institute of Technology Kharagpur, India b School of Medical Science and Technology, Indian Institute of Technology Kharagpur, India c Department of Mechanical Engineering, Indian Institute of Technology Kharagpur, India
Abstract Brain- Computer Interface (BCI) is used to control a system through which people with motor disabilities could achieve a better quality of life by improving their interaction ability with the surrounding environment. Using BCI, patients suffering from severe motor disabilities can also control variety of applications by generating control commands using their own EEG signals. There are many assistive devices are available to reduce the personal, social, and economic burdens of their disabilities and improve their independence but many of these individuals do not have the normal neuromuscular pathway for using their hands to control an assistive device. Hence, EEG could be used to control artificial arm which can help those people to interact with their physical environment and carry out their activity of daily living. In this paper, a genetic algorithm is proposed for trajectory planning of an EEG controlled robotic arm. EEG data for motor imagery were captured from five healthy subjects and left-right hand movement was classified using Support Vector Machine classifier (SVM) with the feature used as Power feature co-efficient and wavelet co-efficient. EEG processing and GA algorithm was developed using Matlab. 2015The TheAuthors. Authors.Published Published Elsevier © 2016 byby Elsevier B.V.B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the Scientific Committee of IHCI 2015. Peer-review under responsibility of the Organizing Committee of IHCI 2015 Keywords: Brain-computer interface (BCI); Motor imagination (MI); Electroencephalography (EEG); Wavelet coefficients; Support Vector Machine (SVM); Genetic Algorithm (GA)
1. Introduction Due to several neuromuscular disorders people can lose their normal neuromuscular pathway through which brain communicates with their external environment. Brain Computer Interface (BCI) is a controlling technique between human and computer through which motor-disabled persons are able to communicate with their surroundings1. BCI records signals directly from the brain and converts them into commands for controlling a device. With combination of a neuro-prosthesis, BCI can be used to replace the lost motor functions with artificially
* Corresponding author. Tel.: +91-3222-282300 E-mail address:
[email protected]
1877-0509 © 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the Organizing Committee of IHCI 2015 doi:10.1016/j.procs.2016.04.080
148
Rinku Roy et al. / Procedia Computer Science 84 (2016) 147 – 151
generated controlled signals. Ideally, a person suffering with motor disabilities can imagine a certain movement which is generally executed instantly by BCI system. The BCI system is used to decode the movement imagination (MI) from the brain signal and a neuro-prosthesis is used to produce the movement. In this paper, EEG signals for left-right arm movement imagination were captured from motor cortex area of the subjects. Power feature co-efficient and wavelet co-efficient were extracted as feature from the acquired EEG signal. Features for the movement imagination are then classified to generate control signal using Support vector machine (SVM) classifier. Genetic algorithm (GA) was used for path planning and to reach the object position. 3 Degree of freedom (DOF) of the human hand was used for all forward and inverse kinematics calculation. Overall process of the experiment is described in Fig. 1.
Fig. 1. Overall process of trajectory planning for EEG controlled arm using GA
2. Methods 2.1 Subjects & EEG recording Five healthy right-handed subjects (4 male and 1 female, mean age: 25 years) participated in this experiment. Before this experiment, none of the subjects were informed about the experimental procedure. C3, C4 and Cz (Fig. 2) electrodes were selected for this experiment as they are nearest to the various motor areas from the scalp. The EEG data were captured using FlexComp Infiniti System (Thought Technology Inc., Canada) with the sampling frequency of 256 Hz. 2.2 Experimental paradigm During experiment, Subjects were seated comfortably in an arm chair in front of a computer screen in which visual cues were shown. They were trained with actual hand movement for 20 trials. After completion of the training procedure, the subjects were instructed to think about their left or right hand movement as per displaying cues.
Fig. 2. Electrode position based on 10/20 system
Fig. 3. Timing details for the visual stimulus
In the instruction set, one cue contains an arrow pointing to the right, indicating right hand motor imagery (MI), other is pointing left, indicating left hand movement imagination. Between each two successive motor imagination task, there was a blank cue was included in the instruction set. At that time subject sit in the resting position, he
Rinku Roy et al. / Procedia Computer Science 84 (2016) 147 – 151
neither moved nor imagined about any body part movements. During starting of the each trial a cross was shown on the computer screen to fixate the subject’s gaze and to suppress eye movements. The exact sequence of a trial is shown in the Fig. 3. We recorded 10 trials per run and 2 runs, in total 20 trials with equally distributed classes. 2.3 Pre-processing of the EEG data: The recorded data were analyzed offline using Matlab. To remove the unwanted artifacts, the signals were filtered with a 5th order Butterworth band pass filter with a cut off frequency of 0.5 to 45 Hz. As movement imagination of the human hand reflected in the Mu (8-12 Hz) and beta (12-30 Hz) frequency band 2, 3, 4; Hence, these two frequency bands were extracted from the EEG data. Imagination of the left hand movement results mu ERD over contra-lateral area and ERS over the ipsi-lateral side as shown in Fig. 4.
Fig. 4. Left hand movement imagination results ERD in contra-lateral side (C4) and ERS in ipsi-lateral side (C3)
2.4 Feature Extraction: Various time domain, frequency domain, and time-frequency domain methods can be used for feature extraction process of the motor imagery EEG5. In this experiment wavelet coefficients and power feature coefficients were extracted for left-right hand movement imagination. x
Wavelet Transformation As EEG signal is non-stationary signal, wavelet transform (WT) was the most appropriate feature to represent the EEG in time-frequency domain. Five levels wavelet decomposition was applied on the raw EEG signal. 4th level detail coefficients were extracted from the filtered EEG as it provided the information of 8Hz-12Hz mu band. x
Power feature ERD/ERS is represented as the change in power of a frequency band (mu/beta band) in EEG signal; so this is a simple method for extracting Event related (de) synchronization. In this method, filtered signals at mu and beta frequency range are squared to get power samples and then average to smooth the data and reduce variability. 2.5 Support Vector Machine (SVM) Classifier:
Various classifiers are used in BCI to generate control signal from the EEG. Here, Support Vector machine (SVM) with 10 fold cross-validation is used to classify left-right hand movement imagination. The support vector machine (SVM)6 uses a linear discriminant hyper plane to identify classes. It maximizes the margin of separation between two classes, i.e., the distance from the nearest training points. 3.
Human Arm Kinematics Kinematics is the study of motion of the object without considering the forces that create the motion. Calculation
149
150
Rinku Roy et al. / Procedia Computer Science 84 (2016) 147 – 151
of the end effecter position of the robot from some specified joint parameters is known as forward kinematics7. On the other hand, inverse kinematics computes joint parameters for a desired end-point position. Kinematic chain of human arm consist of three joints involving spherical joints as shoulder and wrist joint and a hinge joint as the elbow joint. The spherical joints have 3 DOFs while the hinge joint has only 1DOF, which provides a total of 7 DOFs for this human arm kinematics. 4.
Genetic Algorithm (GA) :
GA has been already proposed for trajectory planning using different methods. GA technique was used by P. Garg and M. Kumar8 for identifying the optimal trajectory of robot arm. Pires and Machado9 proposed a GA based path planning using the direct and inverse kinematics. In this experiment, GA was used for path planning of the human arm to reach the goal position. The overall GA process used in path planning of the human hand is described in Fig. 5.
Fig. 5. Path planning of the human arm using GA
5.
Results
The raw EEG data recorded from the subjects were imported in the Matlab (Mathworks Inc. USA) for further off line -processing. After removing the 50Hz noise using a bandpass filter with 0.5 to 45 Hz cut off frequency; power feature coefficients and wavelet coefficients were computed from the filtered EEG. For calculating the wavelet coefficients, Daubechies wavelet of order 3 (db3) was used (Fig. 6(a)). From the calculated coefficients mean, standard deviation and entropy were computed as decoded feature.
Figure 6: (a) Level 4 Db3 wavelet coefficients (b) Power feature coefficients for left and right hand motor imagery movement
In the Fig. 6(b) it is shown that during right and left hand movement power feature coefficients in electrode (C3) and electrode (C4) is easily distinguishable and can be used for classification. Decoded power feature coefficients and wavelet coefficients were classified for left and right arm movement using 10 fold cross validation SVM classifier. The accuracy of the wavelet-SVM and power feature-SVM for the five subjects is shown in the Table 1.
Rinku Roy et al. / Procedia Computer Science 84 (2016) 147 – 151 Table 1. Accuracy for the SVM classifier Features Subject 1
Wavelet coefficient 55.38
Power feature coefficient 59.03
Subject 2
62.1
70.43
Subject 3
63.62
60.02
Subject 4
75.77
65.17
Subject 5
69.93
63.06
Mean
65.36
63.542
GA was using for path planning of the human arm. It randomly generated the population until the end effector had reached to the object. An artificial arm was simulated in Matlab using Genetic algorithm. In the Fig.12 the position of the red object was used as the goal position of the arm. Based on this final position random populations were generated by the genetic algorithm. The trajectory path of the arm to reach the goal position is shown in the Fig 7 with a red line.
Fig. 7. (a) Initial position of the arm; (b) Final position of the arm with the trajectory path planned by GA
6.
Conclusion
In this work, we have presented a trajectory path for an EEG controlled robotic arm. 75.77% classification accuracy is obtained for the left-right arm movement. Between the wavelet coefficient and power feature coefficient, wavelet coefficient provides the better feature classification accuracy. Genetic Algorithm is used for trajectory path planning of 3DoF human hand. In the future work, we could apply GA in human hand kinematics with higher DoF. Thus the above GA based EEG controlled robotic arm could be used for real time applications. References 1.
2. 3. 4.
5. 6. 7. 8. 9.
Laura Kauhanen, Pasi Jylanki, janne Lehtonen, Pekka Rantanen, Hannu Alaranta, and Mikko Sams, "EEG-Based Brain-Computer Interface for Tetraplegics," Hindawi Publishing CorporationComputational Intelligence and Neuroscience, Volume 2007, Article ID 23864, 11pages doi:10.1155/2007/23864 N.E. Crone, D.L. Miglioretti, B. Gordon, et al., Functional mapping of human sensorimotor cortex with electrocorticographic spectral analysis. II. Event-related synchronization in the gamma band. Brain, 121, 2301–2315, (1998). G. Pfurtscheller, C. Neuper, D. Flotzinger, et al., EEG-based discrimination between imagination of right and left hand movement. Electroencephalogr Clin Neurophysiol, 103, 642–651, (1997). L. Leocani, G. Magnani, and G. Comi, Event-related desynchronization during execution, imagination and withholding of movement. In: G. Pfurtscheller and F. Lopes da Silva (Eds.), Event-related desynchronization. Handbook of electroenceph and clinical neurophysiology, vol. 6, Elsevier, pp. 291–301, (1999). N. Birbaumer, “Slow cortical potentials: their origin, meaning and clinical use,”. In G.J.M. van Boxtel and K.B.E. Bcker, (Eds.) “Brain and behavior:past, present, and future,” Tilburg University Press, Tilburg, pp. 25-39,1997. V. N. Vapnik, Statistical learning theory, ser. Adaptive and learning systems for signalprocessing, communications, and control. Wiley, New York,(1998). V. Grecu, N. Dumitru, and L. Grecu,” Analysis of Human Arm Joints and Extension of the Study to Robot Manipulator”, Proceedings of the nternational MultiConference of Engineers and Computer Scientists, Vol 2, March 18 - 20, 2009 pp:1348-1351. Devendra P. Grag, & Manish Kumar, "Optimization Techniques Applied To Multiple Manipulators for Path Planning and Torque Minimization", Engineering Applications of Artificial Intelligence, Volume 15, Number3, June 2002 , pp. 241-252(12). E.J. Solteiro Pires & J.A. Tenreiro Machado, " A Ga Perspective Of The Energy Requirement For Manipulators Maneuvering In A Workspace With Obstacles", Cec 2000-\Congress On Evolutionary Computation , pp. 1110- 1116, 16-19 July 2000, Santiago, California, USA.
151
