*4.1.4. EMG based controlled exoskeleton for hand rehabilitation*

Giuseppina *et al* from Italy have worked on developing a hand exoskeleton system (see Fig. 11) which is also controlled by EMG signals [24]. This is mainly developed for hand rehabili‐ tation where people have partially lost the ability to correctly control their hand movements. Figure 12 shows the control flow chart of the hand rehabilitation robot. During the research they have selected the relevant muscle to capture the finger motion and EMG electrodes are placed in order to minimize the noise due to the movements. This is one of the important factors to be considered when placing dry electrodes on the skin. The control system of the robot consists of microcontroller and EMG acquisition systems. Processed EMG signals are com‐ municated with the microcontroller via serial connection. Finally the microcontroller generates the command signal required to drive the actuator and control real positions by means of sensory inputs. Threshold is defined in order to distinguish the real electric activity of the muscle from other interferences. According to [24], the relationship between motor speed, *v(t)* and joint angle position, *θ (t)* is obtained as in equation (1).

$$w(t) = A\{1 - \Theta(t)\}\tag{1}$$

controller (see Fig. 13). In order to obtain the real time control, a muscle-model-oriented control method has been proposed for the robot. This control method is more suitable compared to the fuzzy-neuro control method, where it needs a higher number of fuzzy rules in case of higher DoF. An impedance controller is applied with a muscle-model-oriented control method and impedance parameters are then adjusted in real time as a function of upper-limb posture and

Patient Electrodes

EMG

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

251

Recent Trends in EMG-Based Control Methods for Assistive Robots

Operator

Interface

Recording & Proceesing

PC

Microcontroller

Actuators Sensors

EMG activity level [6].

Exoskeleton

**Figure 12.** Control flow chart of hand rehabilitation robot [24]

where *A* is defined as an opportunely chosen factor.

According to (1), they are able to control the motor speed which is proportional to the hand opening and progressively it reaches zero when the fingers are flexing. This control method is more suitable for advanced rehabilitation processes. At the same time it takes the effect of natural variability of the EMG signal into account.

#### *4.1.5. Muscle-model-oriented EMG based control of an exoskeleton robot*

The SUEFUL -7 is an upper-limb exoskeleton robot mainly developed to assist motion of physically weak individuals. In the robot EMG signals are used as the main input signal to the

**Figure 12.** Control flow chart of hand rehabilitation robot [24]

**Figure 11.** Hand exoskeleton [24]

250 Electrodiagnosis in New Frontiers of Clinical Research

*4.1.4. EMG based controlled exoskeleton for hand rehabilitation*

and joint angle position, *θ (t)* is obtained as in equation (1).

where *A* is defined as an opportunely chosen factor.

natural variability of the EMG signal into account.

*4.1.5. Muscle-model-oriented EMG based control of an exoskeleton robot*

Giuseppina *et al* from Italy have worked on developing a hand exoskeleton system (see Fig. 11) which is also controlled by EMG signals [24]. This is mainly developed for hand rehabili‐ tation where people have partially lost the ability to correctly control their hand movements. Figure 12 shows the control flow chart of the hand rehabilitation robot. During the research they have selected the relevant muscle to capture the finger motion and EMG electrodes are placed in order to minimize the noise due to the movements. This is one of the important factors to be considered when placing dry electrodes on the skin. The control system of the robot consists of microcontroller and EMG acquisition systems. Processed EMG signals are com‐ municated with the microcontroller via serial connection. Finally the microcontroller generates the command signal required to drive the actuator and control real positions by means of sensory inputs. Threshold is defined in order to distinguish the real electric activity of the muscle from other interferences. According to [24], the relationship between motor speed, *v(t)*

According to (1), they are able to control the motor speed which is proportional to the hand opening and progressively it reaches zero when the fingers are flexing. This control method is more suitable for advanced rehabilitation processes. At the same time it takes the effect of

The SUEFUL -7 is an upper-limb exoskeleton robot mainly developed to assist motion of physically weak individuals. In the robot EMG signals are used as the main input signal to the

*v*(*t*)= *A*{1 - *θ*(*t*)} (1)

controller (see Fig. 13). In order to obtain the real time control, a muscle-model-oriented control method has been proposed for the robot. This control method is more suitable compared to the fuzzy-neuro control method, where it needs a higher number of fuzzy rules in case of higher DoF. An impedance controller is applied with a muscle-model-oriented control method and impedance parameters are then adjusted in real time as a function of upper-limb posture and EMG activity level [6].

The controller of the SUEFUL-7 [6] uses EMG signals of the user as the primary input infor‐ mation. Also, forearm force, hand force and forearm torque are used as subordinate input information for the controller [6]. This hybrid nature of the control method is a guarantee to activate the SUEFUL-7 even with low EMG signal level. On the other hand, when EMG signals are high, the robot is controlled mainly by the EMG signal generated by user motion. The features of the raw EMG signal are extracted through RMS calculation. This RMS of EMG is fed to the controller. In order to identify the 7DoF motions the EMG signals of sixteen locations are measured with surface electrodes.

The first stage is the input signal selection and the second stage is muscle model oriented EMG based impedance control. Proper input information is selected to the controller according to muscle activity levels in the first stage. Depending on the RMS of the EMG signal, muscle model oriented EMG based control or sensor based force control is selected under the second

Recent Trends in EMG-Based Control Methods for Assistive Robots

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

253

Tremor is defined as the involuntary motion that may occur in various parts of the body, such as the leg or arm. An essential tremor is the most common tremor disorder of the arm and it may occur during a voluntary motion such as writing, painting, etc. If the essential tremor occurs in the arm, the person may not be able to achieve sensitive tasks properly with certain

Saga University, Japan has developed an EMG based control method to suppress the tremor of the hand [20]. The features of the EMG signals are extracted from the RMS and fed to the controller. Sixteen EMG channels are selected to measure the muscle activity and they are used to determine the joint torque by knowing weight value for particular EMG signal. This weight depends on upper limb anatomy or result of experiment [20]. The essential tremor is a rhythmic motion and its vibrational component is extracted by using a band pass filter in the controller.

stage and it is fed as a control command to the robot.

Also, the user intention is extracted by using a low pass filter.

*4.1.6. Use of EMG to tremor suppression control*

**Figure 14.** NEUROExos with a subject [10]

tools [20].

The correct joint torque is affected by the posture of the upper limb and it changes the relationship between EMG signals and generated joint torques. Further, this posture variation is nonlinear and stochastic [6]. Fuzzy reasoning is therefore applied to estimate the effect of posture change. Neuro-fuzzy modifier is then applied to modify the muscle model matrix by means of adjusting weights in order to take the effect of changes of posture of the upper limb effectively. Online adaptation of the neuro-fuzzy modifier is important if the robot is expected to be used for different uses. Therefore, the neuro-fuzzy modifier is trained for each user using relevant information. The overall structure of the controller (see Fig. 13) consists of two stages.

**Figure 13.** Structure of SUEFUL-7 controller [6]

The first stage is the input signal selection and the second stage is muscle model oriented EMG based impedance control. Proper input information is selected to the controller according to muscle activity levels in the first stage. Depending on the RMS of the EMG signal, muscle model oriented EMG based control or sensor based force control is selected under the second stage and it is fed as a control command to the robot.

#### *4.1.6. Use of EMG to tremor suppression control*

The controller of the SUEFUL-7 [6] uses EMG signals of the user as the primary input infor‐ mation. Also, forearm force, hand force and forearm torque are used as subordinate input information for the controller [6]. This hybrid nature of the control method is a guarantee to activate the SUEFUL-7 even with low EMG signal level. On the other hand, when EMG signals are high, the robot is controlled mainly by the EMG signal generated by user motion. The features of the raw EMG signal are extracted through RMS calculation. This RMS of EMG is fed to the controller. In order to identify the 7DoF motions the EMG signals of sixteen locations

The correct joint torque is affected by the posture of the upper limb and it changes the relationship between EMG signals and generated joint torques. Further, this posture variation is nonlinear and stochastic [6]. Fuzzy reasoning is therefore applied to estimate the effect of posture change. Neuro-fuzzy modifier is then applied to modify the muscle model matrix by means of adjusting weights in order to take the effect of changes of posture of the upper limb effectively. Online adaptation of the neuro-fuzzy modifier is important if the robot is expected to be used for different uses. Therefore, the neuro-fuzzy modifier is trained for each user using relevant information. The overall structure of the controller (see Fig. 13) consists of two stages.

> Muscle-model-oriented EMG based Impedance control

Sensor-Based force control

Control Command

Input Signal Selection

Shoulder vertical angle Shoulder horizontal anngle Shoulder rotation angle Elbow angle Wrisr angle Wrist force sensor Forearm force sensor Forearm torque sensor

are measured with surface electrodes.

252 Electrodiagnosis in New Frontiers of Clinical Research

EMG RMS-ch1 EMG RMS-ch16 Wrist force sensor Forearm force sensor Forearm torque sensor

First Stage

**Figure 13.** Structure of SUEFUL-7 controller [6]

Second stage

Tremor is defined as the involuntary motion that may occur in various parts of the body, such as the leg or arm. An essential tremor is the most common tremor disorder of the arm and it may occur during a voluntary motion such as writing, painting, etc. If the essential tremor occurs in the arm, the person may not be able to achieve sensitive tasks properly with certain tools [20].

Saga University, Japan has developed an EMG based control method to suppress the tremor of the hand [20]. The features of the EMG signals are extracted from the RMS and fed to the controller. Sixteen EMG channels are selected to measure the muscle activity and they are used to determine the joint torque by knowing weight value for particular EMG signal. This weight depends on upper limb anatomy or result of experiment [20]. The essential tremor is a rhythmic motion and its vibrational component is extracted by using a band pass filter in the controller. Also, the user intention is extracted by using a low pass filter.

**Figure 14.** NEUROExos with a subject [10]

The desired hand position is then obtained considering the summation of the above two amplitudes. Therefore *Xavrg* represents the desired hand motion while suppressing the tremor.

$$\begin{array}{ccccc}\hline \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ } & \text{ }$$

17

255

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

**Figure 15.** Proportional control block diagram of NEUROExos [27].

PA & FA

PA & IA

PA & EA

MA & FA

MA & IA

Controller selection

Forearm motion & wrist flextion/ extension fuzzy-neuro controllers t 1 t 2

t 3 output

Angles of forearm & wrist flextion/ extension

Recent Trends in EMG-Based Control Methods for Assistive Robots

MA & EA

SA & FA

SA & IA

SA & EA

Wrist radial/ulnar fuzzy-neuro controller

EMG RMSs (CH.1-CH.6) Force sensor signals Torque sensor signals

> EMG RMSs (CH.2 - CH.5) Force sensor signals

> > input

**Figure 16.** EMG-based fuzzy-neuro control method [35].

Where *Xavg*, *Xuser* and *Xtre* are desired hand position, rhythmic motion and user intention respectively.

Further, a muscle-model matrix modifier is defined to take the changes of hand posture and minimize the effect of variation of EMG signal and hence torque variation. Also the amount of tremor is not uniform to all; therefore training of the muscle-model matrix is needed prior to use with the subject. Especially, in case of tremor, the training of the muscle-model matrix does not become easy, because the pattern of tremor is not uniform to all.

## *4.1.7. EMG based proportional control method of NEUROExos*

NEUROExos [27] is an upper limb assistive robot designed to be controlled by EMG signals. The robot is shown in Fig. 14. Carrozza *et al* pointed the importance of understanding the accurate torque estimate for assistive robots. They have developed an EMG based proportional control method to estimate the required torque to operate the NEUROExos. [27]. In the control method, Raw EMG signals were processed to obtain the linear envelope (LE) profiles which resemble the muscle tension waveforms during dynamic changes of isometric forces [27]. These LEs were obtained on-line through full-wave rectification of band passed EMG signals and post-filtering by means of a second-order low-pass Butterworth filter with a cut-off frequency of 3 Hz [27].

The block diagram of EMG based proportional control method of NEUROExos is shown in Fig. 15. As in Fig. 15, two proportional controllers, Kbic (gain for bicep) and Ktric (gain for triceps) are set one after other starting from biceps. Both gains are initially set to zero and gradually increased while the subject moves the arm freely. This gain is increased until the subject feels a comfortable level of assistance. Once Kbic is set, the subject is instructed to repeat the same procedure for Ktric too. The experimental results of proportional EMG based control method of the NEUROExos shown in [27] are proved that the exoskeleton provides extra torque indicating effective reduction of effort spent by the subject for movement generation [27]. Therefore, proportional control of the EMG is one appropriate method to estimate the torque required to move the assistive robot. However, the amount of assistance given by the exo‐ skeleton depends on the subject.

The neuro-fuzzy modifier proposed by Kiguchi *et al* [6] can be applied with modifications for effective training for different subjects and hence it may possible to perform a task in minimum time.

#### *4.1.8. EMG based fuzzy-neuro control method of W-EXOS*

The W-EXOS is a 3DOF exoskeleton robot and its control method is an EMG based fuzzy-neuro control. The control method is illustrated in Fig.16. Surface EMG signals of muscles and sensors

**Figure 15.** Proportional control block diagram of NEUROExos [27].

The desired hand position is then obtained considering the summation of the above two amplitudes. Therefore *Xavrg* represents the desired hand motion while suppressing the tremor.

Where *Xavg*, *Xuser* and *Xtre* are desired hand position, rhythmic motion and user intention

Further, a muscle-model matrix modifier is defined to take the changes of hand posture and minimize the effect of variation of EMG signal and hence torque variation. Also the amount of tremor is not uniform to all; therefore training of the muscle-model matrix is needed prior to use with the subject. Especially, in case of tremor, the training of the muscle-model matrix

NEUROExos [27] is an upper limb assistive robot designed to be controlled by EMG signals. The robot is shown in Fig. 14. Carrozza *et al* pointed the importance of understanding the accurate torque estimate for assistive robots. They have developed an EMG based proportional control method to estimate the required torque to operate the NEUROExos. [27]. In the control method, Raw EMG signals were processed to obtain the linear envelope (LE) profiles which resemble the muscle tension waveforms during dynamic changes of isometric forces [27]. These LEs were obtained on-line through full-wave rectification of band passed EMG signals and post-filtering by means of a second-order low-pass Butterworth filter with a cut-off

The block diagram of EMG based proportional control method of NEUROExos is shown in Fig. 15. As in Fig. 15, two proportional controllers, Kbic (gain for bicep) and Ktric (gain for triceps) are set one after other starting from biceps. Both gains are initially set to zero and gradually increased while the subject moves the arm freely. This gain is increased until the subject feels a comfortable level of assistance. Once Kbic is set, the subject is instructed to repeat the same procedure for Ktric too. The experimental results of proportional EMG based control method of the NEUROExos shown in [27] are proved that the exoskeleton provides extra torque indicating effective reduction of effort spent by the subject for movement generation [27]. Therefore, proportional control of the EMG is one appropriate method to estimate the torque required to move the assistive robot. However, the amount of assistance given by the exo‐

The neuro-fuzzy modifier proposed by Kiguchi *et al* [6] can be applied with modifications for effective training for different subjects and hence it may possible to perform a task in minimum

The W-EXOS is a 3DOF exoskeleton robot and its control method is an EMG based fuzzy-neuro control. The control method is illustrated in Fig.16. Surface EMG signals of muscles and sensors

does not become easy, because the pattern of tremor is not uniform to all.

*4.1.7. EMG based proportional control method of NEUROExos*

respectively.

254 Electrodiagnosis in New Frontiers of Clinical Research

frequency of 3 Hz [27].

skeleton depends on the subject.

*4.1.8. EMG based fuzzy-neuro control method of W-EXOS*

time.

*X avg*<sup>=</sup> *Xusr*- *Xtre* (2)

**Figure 16.** EMG-based fuzzy-neuro control method [35].

of the exoskeleton (hand force and forearm torque) robot [35] are used as input information to the controller. This fuzzy-neuro control method consists of a flexible fuzzy control and adaptive neural network control which is used to obtain natural and flexible motion assist. Fuzzy if-then rules have been constructed to determine the required torque to the motor according to the motion intention of the human. In total, nine fuzzy-neuro controllers are used and this allows operating the exoskeleton robot flexible with EMG signals. Depending on the subject and nature of power-assistance, training of the fuzzy-neuro controller is performed.

The main drawback of this kind of control method is the difficulty of defining the fuzzy if-then control rules when the controller is applied for exoskeleton robots with higher DoF. Further, training of the controller is essential even when the physical and psychological conditions change in the user.
