*4.2.5. EMG based ANN controller for a transhumeral prosthesis*

bit has defined three digital levels. Accordingly, an input comprises of three muscle contrac‐ tions to generate three EMG levels and 27 different commands. From the practical point of view, 18 out of 27 commands were selected for the implementation. However, no information relevant to the pattern recognition has been published by the authors. Table 3 shows the main commands of the 3-bit command language extracted from [45]. According to the controller mechanism no simultaneous operation could be realized using the MANUS hand. Therefore instead of the 3-bit command language it would be better to use a proper pattern recognition method with a higher number of EMG inputs. This could also be coupled with an EEG or other

Disjoint Windowing

Front-end Processing

Bayesian Fusion Majority Vote

**Figure 18.** Block diagram of experimental evaluation of the EMG-pattern recognition system [46]

Evaluation Postprocessing Classification

Overlapping Windowing TD+AR+Hjorth/LDA

LIBSVM kNN

Feature Extraction/Re duction

biological signal, in order to achieve higher effectiveness from the controller.

Data Collection

260 Electrodiagnosis in New Frontiers of Clinical Research

Performance

**Figure 19.** MANUS hand [23]

The EMG based Artificial Neural Network (ANN) controller was developed to realize elbow flexion/extension and forearm supination/pronation based on the artificial neural network (ANN) techniques using EMG signals as the input. Fig. 20 shows the basic concept of the design of the controller. EMG signals are fed into the controller from seven muscles: brachialis, biceps, medial head of triceps, posterior deltoid, anterior deltoid, clavicular and pectoralis major. Raw EMG signals from the muscles are amplified, alternating current coupled, low-pass filtered and recorded. Recorded EMG signals are processed offline. They are filtered, windowed and features extracted. Several features are extracted from 320 samples, 128ms sample time rectangular windows with 50 percent overlap between adjacent segments which provides a sample time of 64ms. The features, mean absolute value, waveform length, number of zero crossing and number of slope sign changes are extracted from each window, which generate a four-element feature set for each EMG channel. Simultaneously motion capturing data from the Optotrak Certus motion capture system is resampled such that its sample time matched that of the EMG data and average joint angles were extracted.

attached to the prosthetic socket as shown in Fig. 21. EMG data and digital information transmit between the implants and the telemetry controller forming a magnetic link through the coil. The data is then converted into analog form at the controller and can be used to control the prosthetic device. When the results from iEMG are compared with sEMG for the same user, iEMG results show a drastic improvement in providing the same motion of the natural hand. Even though iEMG is capable of providing better signals, it is an invasive process and may cause discomfort to the user. Therefore, a hybrid control method, EMG coupled with EEG would result in providing a better control for the prosthetic device. Further, it will not cause

Recent Trends in EMG-Based Control Methods for Assistive Robots

http://dx.doi.org/10.5772/56174

263

In this chapter, recent EMG based control methods of assistive robots were comprehensively reviewed. As assistive robots, upper-limb exoskeleton robots and upper-limb prostheses were mainly considered here. At first, the detection and processing of the EMG was explained discussing available EMG extraction systems. Then ways of categorization of EMG based control methods as input information to the controller, architecture of the control algorithm, output of the controller and other ways were discussed. EMG based control methods were categorized based on structure of the control algorithm as pattern recognition based control and non-pattern recognition based control. Even though assistive robots with pattern recog‐ nition based control can be commonly found, control methods that belong to non-pattern recognition based control are hardly found. Recent EMG based control methods of upper-limb exoskeleton robots and prostheses were reviewed separately. In the review recent EMG based control methods of upper-limb exoskeleton robots and upper-limb prostheses were compared

In addition to EMG signals, EEG, EOG and MMG signals also represent the human motion intention and these can be used as input signals to the controller of the assistive robots. Accordingly, in the future a hybrid control algorithm can be developed with a combination of two or more biological signals as inputs to the controller. Assistive robots are expected to function and appear as their biological counterparts. That is, an exoskeleton should ideally act as a second skin for the human and a prosthetic device the same as the natural limb. Accord‐

Actuator technology also plays an important role of control methods of exoskeleton robot and prostheses. Actuators drive a robot interacting with human according to control inputs. Linearity, stability, correspondence between human motion and actuator actions are important terms for successful function of actuators and hence the control system too. New actuator technologies can be successfully used with exoskeleton robots and prostheses to improve its

In future developments, the aspect of control systems of the assistive robot can be extended to take the effect of microclimate conditions present around the user and take suitable control efforts to provide comfort to the user. Since exoskeleton robots and prostheses closely interact

ingly, control methods of devices should be improved in the future.

discomfort to the user.

considering their features.

function.

**5. Conclusion and future directions**

**Figure 20.** Block diagram of control strategy [36]

**Figure 21.** Schematic of the implantable myoelectric sensor system[8]
