A Synergetic Brain-Machine Interfacing Paradigm for Multi-DOF Robot Control
Summary (4 min read)
Introduction
- Depending on the nature of the experiment, the acquired EEG is found to have specific signal characteristics.
- Study on the co-adaptivity of the user with the BMI system is an active area of research and to date, there is not much literature available on auto-adaptive and autocalibrated approaches.
- As a result, such control techniques do not provide a practical solution and are far from natural human limb coordination since it is ideal to employ a control framework which allows users to drive BMI-driven robot as a third arm by his feeling.
- This section also provides information on the experimental setup.
A. Synergetic BMI Control Scheme
- It is known that human beings do not perform the joint actions of compound movements consciously.
- But on gradual and repetitive trials of the same movement, the cerebellum begins to take control of the task by recognizing the relation to each segment of consciously initiated movement.
- The aim of BMI control of a prosthetic or robotic limb is to allow seamless human-like movement but to date, they incur joint redundancy issues during movement tasks.
- First, the decoder/ classifier is designed to continuously adapt to the changing brain signal, while the subject simultaneously observes the movement of the robot.
- Because of the two adaptive function, the subject is free to control the robot arm without burdening himself to control complex joint management.
B. Experiment Description
- In the first day, the subjects perform the tasks on two separate sessions.
- The data from the first session is used to train the decoder, while the same from the second session is used for offline testing the training of the decoder.
- Fig. 3 shows the generic structure of the visual cue.
- Here, the online task required the subject to guide the robot end-point toward the target based on the instructions from the operator.
C. Co-Adaptive EEG-BMI System
- The BMI system employs wavelet transforms [40], [41] for feature extraction, Laplacian EigenMaps [42] to determine the relevant features and an SVM classifier [43] to decode between the two mental states.
- The filtered signals are then processed using discrete wavelet transform (DWT) [41] to derive the signature features related to left- and right-MI.
- The authors have determined the optimal dimensionality of relevant features for each subject from their validation results.
- The aim of the SVM classifiers is to determine the separating hyperplane with the maximum margin.
- 2) Input N − L datapoints to the trained decoder and determine their respective posterior probabilities (P).
D. Peripheric Motor Learning of the Dynamic Environment
- Tacit learning employs the command signal accumulated during repetitive interaction with the environment to develop an appropriate behavior for the system.
- 2) The FB force error is mapped into the joint torque space by using the Jacobian of the robot arm and the motorcommand error works as a supervising signal.
- Thus, each joint has a local torque control to generate the specified joint torque for the robot.
- The control algorithms are executed on a master PC with the interface of analog-todigital and digital-to-analog converters from the encoders and to the motors, respectively.
- The configuration allows the joints to be controlled independently and thus it can be presumed as a modular structure present within cerebellar pathways.
E. BMI Evaluation Metrics
- To evaluate the performance of the BMI system during training and validation, the authors have employed four quantitative measures.
- It is the measure of how correctly a classifier has classified the positive class.
- Thus, the authors can say ROC curve is a plot of the classification result of the most positive classification to the most negative classification and the resultant AUC is widely used as a classification metric.
- The authors have quantified the performance of the online task of moving the robot arm using left and right hand MI by the following metrics: 1) accuracy and 2) time taken, i.e., the time taken to process and decode the incoming EEG signal and transmit it remotely to the robot using SSH protocol.
III. RESULTS
- This section begins with the detection of ERD/ERS signals from the EEG acquired from the Emotiv system.
- Then, it presents the results on the performance of the BMI system during training and offline testing of the decoder, performance of the peripheral motor controller during its learning stage and the complete performance of BMI system (which includes the trained BMI decoder and the trained peripheral motor controller) during online experimentation.
- The offline processing and online experimentation has been executed in MATLAB Windows 8.1 environment.
A. Detection of ERD/ERS Patterns
- The Emotiv acquisition system, used in this paper, does not have any channels directly over the primary motor cortex, but it has channels, FC5, FC6, P7, and P8, in the vicinity of the region.
- Thus, for MI studies, one can use these channels to detect the ERD/ERS waveform.
- Hurtado-Rincon et al. [52] and Dharmasena et al. [53] has successfully classified between TABLE II VALIDATION OF THE DECODER ON A NEW DATASET (OF 40 TRIALS) DURING NO ADAPTATION AND ADAPTATION left- and right-MI.
- Fok et al. [54] have acquired brain signals related to movement to successfully drive a powered orthosis tasked at opening and closing of the patient’s hand.
- As noted from the plots, the right side of the brain is more active during left hand MI and vice-versa.
B. BMI Adaptive Decoder Training and Validation
- In Table I, the authors have shown the average of the classification metric [i.e., accuracy, sensitivity, specificity, and AUC (in %)] for all the nine subjects.
- The adaptation result suggests an increase of average accuracy, sensitivity, specificity, and AUC by 9.03%, 7.54%, 8.77%, and 6.37%, respectively, from its nonadaptive counterpart.
- The p-values as observed from Table I suggests that for all subjects rejects the null hypothesis at 5% significance level, and thus, it is statistically shown that the adaptive decoder is more accurate than the nonadaptive one.
- The average sensitivity, specificity and AUC being more than 85% suggests that the decoder can detect 85% of the positive and negative classes without adversely affecting each other.
- The positive result shown during validation allowed us to use the decoder during online testing of the BMI system.
C. Learning of the Synergetic Motor Controller
- For this experiment, the authors have used two different loads of 300 and 600 g as unknown loads for the robot.
- It means the load is not a-priori known for the motor controller.
- Fig. 8(a) and (b) illustrates the trajectory of the robot arm during its learning for both the loads.
- The time sequential transition of the end-point of the robot for both the figures are illustrated using a color map which changes with the progress of time.
- Fig. 8(c) shows the shoulder-elbow phase map for the different weights.
D. Online Performance of the Simultaneous Multi-DOF Robot Control by Co-Adaptive BMI
- Following the training of robot controller using synergetic motor learning algorithm, the subject is ready to move the robot arm by his/her motor intention.
- The decoder decodes the brain signal to generate the corresponding control command necessary to move the robot in either up or down direction.
- As seen from the figure, the robot requires a number of steps (or trials of MI extraction) to reach the target.
- This observation is also validated by the joint angle variance metric shown in Table III.
- This observation regarding minimal shoulder and wrist usage for heavy object manipulation is well matched to the situation in human motor control.
IV. DISCUSSION
- The authors describe some co-adaptive approaches implemented by other researchers.
- In another interesting work, Kus et al. [57] developed a BCI system which followed an asynchronous mode of operation, automatic selection of parameters based on initial calibration and incremental update of the classifier parameters from FB.
- The participants performed right hand, left hand and foot MI based on instructions from a visual cue with an accuracy of 74.84%.
- The authors have employed this form of adaptation to the subject to make the task more realistic and practical.
- The advantage of a separate motor learning control scheme, even for 3-DOF joint control, allows the subject to focus on the lower dimensional endpoint control of the robot while the proprioceptive information from the robot is processed inside the peripheral motor control and adapts accordingly while performing simultaneous multijoint control.
V. CONCLUSION
- The authors have proposed a new BMI paradigm which integrates an MI EEG to extract the target intention with adaptive decoder for cortical signals and a synergetic motor learning control to cope with the peripheral control of a multijoint redundant robot arm with environmental dynamics adaptation capability.
- The proposed method allowed for BMIcontrolled robot to employ different joint usage depending on the given payload systematically through the learning process.
- To the best of the authors’ knowledge, it is a first system which incorporates dual adaptive nature in each cortical level and peripheral motor control level in BMI.
- The positive result, thus obtained, has opened a door to proceed forward in this research, but it was verified with simple task as a starting point.
- To improve the speed and robustness of the BCI control alogrithm, the authors would design a self-paced experiment with the provision of an error FB through EEG [15].
Did you find this useful? Give us your feedback
Citations
26 citations
Cites methods from "A Synergetic Brain-Machine Interfac..."
...Those were achieved using different BCI signals like SSVEP [53] MI [13, 54]....
[...]
...Motor imagery is used in most of the position control applications to actuate the external device based on right/left hand-motor imagination [12-13]....
[...]
22 citations
19 citations
16 citations
16 citations
Cites background from "A Synergetic Brain-Machine Interfac..."
...proposed in our previous study of the coordination of a redundant robotic arm [23], [24]....
[...]
References
19,398 citations
7,210 citations
6,803 citations
4,226 citations
"A Synergetic Brain-Machine Interfac..." refers background in this paper
...ally defined as “minimum X,” where X is jerk [31], torque...
[...]
2,469 citations
"A Synergetic Brain-Machine Interfac..." refers background in this paper
...Finally, the cerebullum attains control over the entire process and by a mere trigger from the cerebrum, it can execute the entire movement without any conscious effort [28]–[30]....
[...]