**4.1. EMG based control methods of upper-limb exoskeleton robots**

Different approaches have been proposed by various researchers in the past in order to estimate the muscle torque starting from EMG activation. These methods include neural networks, neuro-fuzzy classifier, hill model *etc* [27]. In the next subsections, several EMG based control methods of upper-limb exoskeleton robots are reviewed.

### *4.1.1. EMG based control of hand exoskeleton*

The hand exoskeleton robot with EMG control has been developed by researchers from the University of Berlin, Germany [26]. This mainly focuses to use by patients who have limited hand mobility. Figure 8 shows the EMG control method for a hand exoskeleton with blind source separation. The construction of the design allows controlling the motion of finger joints. Researchers have presented the difficulties in the application of EMG algorithms. One such drawback is the identification of the correct muscle responsible for a particular motion. In other words, only a subset of muscles responsible for hand movement is sampled by the surface electrodes [26]. Another difficulty is the EMG signal separation for relevant motion. This needs a suitable process to recover the underlying original signals. Armin *et. al.,* have proposed a blind source separation method to recover the underlying original signals developed by a particular motion of muscle [26]. Initially signals are filtered by a weighted low pass differ‐ ential filter. Then an inverse demixing matrix which, results from an iterative algorithm [4] approximates the separation about 1.5dB for close proximity sensors. Subsequent to the separation, the signals are used for control purposes. However, integration of additional sensors and additional DoF complicates the separation and further, the position of electrodes increases the complexity. Therefore, blind source separation has practical limitations in working with EMG sensors with force sensors.

*4.1.2. EMG based neuro-fuzzy control method*

**Figure 8.** Structure of the control method with blind source separation [26]

Blind Source Seperation

Thumb flexor/ extensor

Kiguchi *et al*. [18] have developed an exoskeleton robot and it is controlled based on EMG signals. The robot is used to assist the motion of physically weak persons such as elderly, disabled and injured [37-39]. Although EMG signals directly reflect the human motion intention, it is difficult to control the robotic exoskeleton since the strength of EMG varies with factors like physical and physiological conditions, placement of electrodes, shift of electrodes and high nonlinearity of muscle activity for a certain motion. Therefore, Kiguchi *et al* proposed a neuro-fuzzy controller with EMG signals which provides flexible and adaptive nonlinear control for the exoskeleton robot in real time. The architecture of the controller is shown in the Fig. 9. Mean absolute value (MAV) is used to extract the features of the EMG signal due to its

Threshold Threshold Threshold Threshold

Motor Control

EMG Aquisition

Blind Source Seperation

Rectification, Low pass filtering

Decomposition

Middle finger flexor/extnsor

Index finger flexor/ extensor

Blind Source Seperation

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

247

Recent Trends in EMG-Based Control Methods for Assistive Robots

Right/little finger flexor/extensor

**Figure 8.** Structure of the control method with blind source separation [26]
