NEUROLAB: A MULTIMODAL NETWORKED EXOSKELETON

FOR NEUROMOTOR AND BIOMECHANICAL RESEARCH

A. F. Ruiz, E. Rocon, F. Brunetti, L. Bueno, J. C. Moreno and J. L. Pons

Bioengineering Group, Instituto de Automatica Industrial, Ctra. Campo Real km 0.200, Madrid, Spain

Keywords:

Biorobotics, Biomechanical Devices, Exoskeletons, Health Monitoring Devices.

Abstract:

NeuroLab refers to an experimental platform designed to enhance studies in human movement and neuro-

motor control. The platform comprises a robotic exoskeleton and some other stand-alone devices. All of

these components have communication capabilities integrated in hardware and can work cooperatively taking

advantage of a networked architecture. A set of experiments have been conducted with NeuroLab. The

objective of the trials was to use mechanical perturbations to identify the viscous-elastic properties in human

elbow joint and to correlate such mechanical impedance with the electromyographic information of muscles

associated to the joint, during a postural task and in a rest position. In each condition, a pseudo-random torque

perturbation was applied directly to the arm and to the forearm by mean of an upper limb powered exoskeleton.

The angular kinematics (velocity and position), kinetics (torque) and the muscular activation patterns (EMG)

in the two main muscles (biceps and triceps brachii) intervening in the elbow ﬂexion-extension movement

were recorded.

1 INTRODUCTION

Human movement and neuro-motor control is a very

complex research ﬁeld due mainly to the complexity

of the involved mechanisms and the difﬁcult access to

the components of the overall system. Due to these

reasons, the research community tries to exploit all

kinds of valid information (EMG, EEG, kinetics and

kinematics) relating to movement planning and exe-

cution in order to understand this complex system and

to develop new aids in the medical robotics ﬁeld.

One common and generally-accepted approach to

understanding and modelling the human motor sys-

tem is to monitor and analyse movement-related data

during different motor task. A common approach to

understand the dynamics of the motor control system

is to independently manipulate the mechanical condi-

tions of each joint while acquiring the biomechanical

signals and the generated bio-potentials while the hu-

man motor system adapts to those new applied condi-

tions.

In this scenario, a set of tools attached to the hu-

man body is required. In NeuroLab, there are inde-

pendent devices that communicate with each other,

based on a Personal Area Network (PAN) concept.

Each device has a speciﬁc function and helps to ad-

dress the overall goal of the platform. The NeuroLab

integrates several devices in a global architecture. The

main goals of NeuroLab are:

1. Study of human movement in subjects with mo-

tor disorders such as pathological tremor or spas-

ticity. The information provided by the platform

during the execution of speciﬁc motor tasks can be

used as a tool to diagnose and assess motor disor-

ders, (Rocon et al., 2007).

2. Study of neuro-adaptative strategies for learning

and training of speciﬁc motor patterns through the

application of selected force-ﬁelds to the upper

limb. This application could potentially be of con-

siderable impact in patients suffering of cerebral

injuries, (Krebs et al., 1998).

3. Validation of neurophysiological models of hu-

man motor control in upper and lower limbs. This

will help to gain a better understanding of the inte-

gration of the sensory information and the under-

F. Ruiz A., Rocon E., Brunetti F., Bueno L., C. Moreno J. and L. Pons J. (2008).

NEUROLAB: A MULTIMODAL NETWORKED EXOSKELETON FOR NEUROMOTOR AND BIOMECHANICAL RESEARCH.

In Proceedings of the First International Conference on Biomedical Electronics and Devices, pages 68-73

DOI: 10.5220/0001049400680073

 SciTePress

lying mechanisms for generation of motor com-

mands.

4. Study of human body behaviour under external

loads. The load application is the basis for several

technical aids to compensate functional disability.

5. Exploration of new communication channels in

human-robot interfaces. This is potentially feasi-

ble through the use of EMG and EEG information

to control wearable robots, (Rosen et al., 2001),

(Pfurtscheller et al., 2002).

6. Assessment and quantiﬁcation of human upper

limb parameters, e.g. mechanical impedance.

These parameters are considered important for

understanding of the control mechanisms of the

human joints, the generation of control signals,

the execution of movements and the adaptation

under changing conditions.

This paper aims at describing the design and the

development of a platform to enhance research in se-

veral ﬁelds. The next section presents each device

concept of the NeuroLab system. Next, a set of ex-

periments which are being conducted with such plat-

form in order to model the human motor control at the

upper limb will be described. The experimental me-

thods and preliminary results are presented in section

3. Finally, the section 4 discusses future work with

NeuroLab.

2 PLATFORM DESCRIPTION

The platform (see ﬁgure 1), is composed of modules

and devices that provide several capabilities: an upper

limb robotic exoskeleton, an EMG module, a Biome-

chanical Monitoring module, and an EEG module. It

can further be expanded with other peripherals. A

software platform is deﬁned to manage the system,

e.g. setup the experiments and acquire data. Safety

and reliability were priority considerations in the de-

velopment.

The powered exoskeleton and the devices can

communicate with each other using a CAN-based net-

work and speciﬁc protocols. Each element of the plat-

form providesseveralservices which can be requested

by other devices. There are therefore different prim-

itives in the upper layers of the protocol, for instance

to retrieve the data acquired by a module or to control

a joint of the exoskeleton.

The robotic device is an upper limb exoskeleton

which allows the mechanical conditions of each limb

joint to be manipulated independently, (Ruiz et al.,

2006). The networked platform enables combined

Sensors

Actuators

Roboticexoskeleton

EMGModule

EEGModule

Biomechanical

Module

Matlab

Bluetooth

Ethernet

PC/104

ECAN

Wired

Others

Central

Platform

Figure 1: Layout of the NeuroLab system. Different mod-

ules can connect to a base station using Bluetooth. At the

same time, all the modules are part of the wearable-robot

network, called BioNET.

measurement of biomechanical variables (kinemat-

ics and kinetics variables) and biopotentials, such as

electromyography (EMG) and electroencephalogra-

phy (EEG).

2.1 Robotic Exoskeleton

The upper limb robotic exoskeleton in NeuroLab

spans the human elbow and wrist joints, (Rocon et al.,

2007). The sensors (gyroscopes, potentiometers and

force sensors) measure the biomechanics of the arm.

Using this data, limb movements, motor tasks and

several postures can be assessed under different me-

chanical conditions.

Maxon Motor EC 45 Flat continuous current have

been selected as actuation device, which is a very

light, small DC motor without brushes that adapts to

orthotic applications. In order to match the speed and

the torque of the DC motor to the application require-

ments, a gearbox was necessary for the system. This

was done via a harmonic drive. In particular, the drive

selected for the application was the HDF-014-100-

2A. The actuator system conﬁgured in this way can

apply a maximum torque of 8 N.m.

The exoskeleton is controlled following an

impedance control strategy which includes a position

feedback loop. The goal of the controller is to modify

the apparent Human–Robot impedance.

NeuroLab has a real-time target computer system

(xPC Target) to control the exoskeleton. Control is

implemented using the MatLab Real-Time suite by

NEUROLAB: A MULTIMODAL NETWORKED EXOSKELETON FOR NEUROMOTOR AND BIOMECHANICAL

RESEARCH

Figure 2: Upper limb robotic exoskeleton. The device spans

the human elbow and wrist joints.

MathWorks, Inc. This environment provides mathe-

matical libraries making it easy to implement control

strategies. The algorithm can be coded in C-language

and compiled in an executable application.

2.2 EMG Module

Measurements supplied by electromyography (EMG)

provide a valuable information regarding physiology

and muscle activation patterns. This information de-

scribes the forces that will be generated by the mus-

cles and the timing patterns of the motor commands.

It can be also used to assess the response of the hu-

man motor system to external dynamic conditions or

perturbations.

The EMG module allows for acquisition of data

on four muscle groups. Since the EMG signal is very

small (50µV- 5mV), it may be affected by interfer-

ence from other biological and environmental noise

sources, e.g. movement artifacts, electric noise and

muscle noise among others, (DeLuca, 1997). In or-

der to minimise the effects of noise, the EMG module

ampliﬁes and ﬁlters the raw EMG signals before they

are digitalized.

Additionally, a battery is used to power the EMG

acquisition module in order to reduce 50 Hz har-

monics (power-line noise). In the light of interna-

tional safety regulations regarding electronic devices

connected to human beings, several topics were ad-

dressed in connection with electric isolation of the

EMG module. In particular, galvanic isolation using a

wide-band, unity-gain isolation ampliﬁer was imple-

mented in the EMG Module.

2.3 Biomechanical Monitoring Module

This module uses inertial sensors to acquire kinematic

and kinetic information on the system. This was the

ﬁrst smart module developed, so further details are

given. The modular approach of NeuroLab enables

the use of the different devices in many different ap-

plications.

The Biomechanical monitoring comprises the fol-

lowing logical components:

• The controller. This uses a TMS320F2812 DSP

Texas Instrument, which is powerful enough to

run all the signal processing algorithms. The

clock frequency is up to 150 MHz. The DSP in-

cludes several communication interfaces.

• The Sensor Set. Two inertial sensors can be con-

nected to the controller using a SPI interface.

Each sensor consists of a set of three gyroscopes,

three accelerometers and three orthogonally-

mounted magnetometers (see ﬁgure 3).

• The Data Logging block. An ATMega32 micro-

controller is used to manage a SD card. The

microcontroller implements a FAT16 ﬁle system.

Using basic commands, the controller can store

the data of the sensors in a non-volatile memory.

• Communication block. The communication block

includes four different communication interfaces

for networks. The ﬁrst is the SPI, which is em-

bedded in the DSP and is used to communicate

with the sensors and the data logger. The second

block comprises a Bluetooth module for wireless

communication with a base station for real-time

monitoring. The third interface is a CAN port pro-

vided by the DSP. It can be attached to the Neuro-

Lab BioNET using simple CAN drivers. The last

interface is an USB port for data transfer and real-

time monitoring. The Biomechanical Monitoring

Module can be connected to the central platform

(Figure 1) using Bluetooth or USB.

• Power supply. This is based on an Ion-Lithium

battery with a capacity of 900 mAh. The module

uses the USB connection to charge this battery.

2.4 EEG Monitoring Module

EEG can be used to study movement planning and to

control wearable robots, (Wolpaw et al., 2002). The

development of portable EEG module for research

purposes is not a trivial task. Noisy environments

and movement artifacts affect the quality of the EEG

signals. Moreover, EEG signal processing techniques

are usually complex and require a powerful platform

BIODEVICES 2008 - International Conference on Biomedical Electronics and Devices

Figure 3: Inertial sensor that include 3 gyroscopes, 3 ac-

celerometers and 3 magnetometers.

to execute these algorithms. Even relatively simple

algorithms can require a powerful platform as usually

the EEG is acquired using arrays with more than 10

electrodes.

Early developments of EEG Monitoring Module

have been based on a PC/104 computer platform. A

special ampliﬁcation board was designed following

the safety requirements for devices directly connected

to the human body. This board is aimed to amplify

16 channels with variable gain ampliﬁers and a band-

width that spans from 0.1 to 80 Hz. The board also

has a notch ﬁlter centered on 50 Hz to compensate

for 50 Hz line noise. This board is connected to the

PC/104 through a data acquisition board. The acqui-

sition of the EEG is aimed to use a 512 Hz frequency,

minimizing distortion due to acquisition. The CAN

connectivity is achieved through the use of an exter-

nal CAN board attached to the PC/104 platform.

The EEG monitoring module can provide EEG

logging through the use of a hard drive connected

to the PC/104 board. Other services of the module

comprises the identiﬁcation of patterns related to the

movement planning and imagination to be used as a

control signal to the robot. This module also uses the

xPC target platform (from MatLab, Inc.) used to con-

trol the robotic exoskeleton. This software platform

was chosen due to the mathematical tools already im-

plemented and for its ﬂexibility.

2.5 BioNET

The purpose of NeuroLab is to integrate several dif-

ferent devices in order to study Human-Robot inter-

action (both cognitive and physical) and the human

neuromotor system using non-invasive techniques. In

view of the wide range of proﬁles and applications of

the system, a distributed modular approach was se-

lected to implement the proposed concepts.

A network of smart devices was identiﬁed as the

optimum solution to achieve the goal. The network is

called BioNET and is CAN-based. The work package

includes the development of several network proto-

cols including service discovery, synchronisation, and

priority management mechanisms among others. A

table describing the device, its services and its param-

eters, is stored in the device itself. This concept is

similar to TEDS, used in IEEE P1451.3.

Current research efforts are aimed to develop the

monitoring and rehabilitation proﬁles for the network.

3 EXPERIMENTAL METHODS

Many studies have approximated the dynamic be-

haviour of human body segments such as upper

and lower limbs and their joints as a mechanical

impedance, (Hogan, 1984), (Dolan et al., 1993),

(Tsuji et al., 1995), (Zhang and Rymer, 1997). The

mechanical impedance in this context can be deﬁned

as the dynamic relation between small force and po-

sition variations.

Using the platform described, NeuroLab, a set of

experiments are being conducted to estimate the prop-

erties of the human elbow joint impedance and to de-

termine viscoelasticity–EMG relationships. This is

supported for the fact that the EMG information can

be also used to assess the response of the human mo-

tor system to external dynamic conditions or pertur-

bations. In literature, several studies have used elec-

tromyography in biomechanical analysis and human

joint torque estimation, (Clancy and Hogan, 1997).

To start with experiments on this topic, a system

for measuring arm impedance is required. Thus, the

robotic exoskeleton is set up as a mechanical mea-

surement system to get reference measurements for

correlation with EMG–signals. The robotic device

applies torque perturbations to the subject’s arm. Sen-

sors of robotic device deliver the necessary data to

compute the mechanical impedance.

The human arm and their articulations could be

modelled as a mechanical impedance in terms of iner-

tia (I), viscosity (B) and elastic stiffness (K), using a

linear second order model (Equation 1), (Dolan et al.,

1993).

The parameters in the model that represent the dy-

namic behaviour of the human neuromusculoskele-

tal system are non-linear and vary highly depending

on factors such as torque bias and posture, (Kear-

ney and Hunter, 1990). Therefore, experiments that

ﬁt the data to an impedance of a second-order linear-

model must specify an operating point. The operating

point consists of constant posture, constant force, and

non-fatiguing contractions over a particular task. The

ensemble of linear models estimated over a range of

operating conditions may be thought of as deﬁning a

quasistatic model of arm dynamics and can be deﬁned

by the following linear equation:

NEUROLAB: A MULTIMODAL NETWORKED EXOSKELETON FOR NEUROMOTOR AND BIOMECHANICAL

RESEARCH

F(t) = I

∂

X(t)

∂

+ B(δ)

∂X(t)

∂t

+ K(δ)X(t) (1)

where F(t) and X(t) represent the force and the dis-

placement, respectively, and δ deﬁnes the operating

point of the system.

According to Equation 1, inertial component re-

main constant and viscous and stiffness components

(B and K) are functions.

3.1 Protocol

Four healthy subjects participated in the experi-

ments. Subjects were instrumented with surface EMG

electrodes according to the SENIAM recommenda-

tions, (http://www.seniam.org). Two muscles agonist-

antagonist involved in the elbow joint movement were

measured: the ﬂexor (biceps brachii) and extensor

(triceps brachii long head) muscles.

Subjects wore a robotic exoskeleton on its right

arm allowing elbow ﬂexion and extension in the ver-

tical plane. Shaft joint on the device was aligned with

subject elbow joint, and the device was attached to its

upper arm and forearm. The elbow was ﬂexed making

an angle of 90 degrees.

The trials consisted of an intentional postural task.

In each trial a pseudo-randomtorque perturbation was

applied directly to arm and forearm by the upper limb

powered exoskeleton.

The duration of each trial was 10 seconds. The

subject was asked to maintain the position while the

mechanical perturbation was applied. Three repeti-

tions were chosen for each experimental session and

the signals were sampled at 1 kHz for biomechanical

variables (kinetics and kinematics) and for the elec-

tromyographic signals (sEMG).

3.2 Data Analysis

Kinematics and kinetics data were ﬁltered using a 4

order Butterworth low-pass ﬁlter with a cut-off fre-

quency of 10 Hz.

The toolbox System Identiﬁcation Toolbox of Mat-

lab have been used to accomplish the modelling pro-

cess. In particular, the function armax was used to

ﬁts the parameters of the linear second-order model

to the structure of ARMAX (Auto-Regressive Moving

Average with eXogenous inputs), based on a predic-

tion error method.

Surface EMG signals were rectiﬁed (full-wave)

and the envelope of the signals extracted using a low-

pass ﬁlter with a cut-off frequency of 10 Hz. A 5

order Butterworth ﬁlter for this purpose was adopted.

The RMS (Root Mean Square) value was used as in-

dex to quantify amplitude of EMG signals as deﬁned

by Equation 2. In the correlation procedure, the RMS

value was the considered variable.

RMS =

∑

i=1

(2)

where x

is value voltage in i

sample, n is number of

samples in segment.

The RMS value represent the root square of the

mean power of the EMG signal for a speciﬁc time

period.

The linear equation that relates EMG amplitude

and the variation of angular position to the variation

in the generated torque by the joint might be modelled

as Equation 3.

∆T = I · ∆

θ+ B( ˆs

, ˆs

) · ∆

θ+ K( ˆs

, ˆs

) · ∆θ (3)

where ˆs

and ˆs

are the amplitude estimation of EMG

signals for muscles ﬂexor and extensor, respectively.

∆θ is the variation in angular position and ∆T is the

variation of torque generated by the joint.

3.3 Results

Figure 4 represents the estimated parameters of me-

chanical impedance and its mean and standard devia-

tion for one subject. Each sample of x-axis in ﬁgure

represents a trial, in order to evaluate the repeatabil-

ity. Each trial magnitude was the mean of estimated

values of a set of two-second windows of the recorded

data.

Several quantitative information have been re-

ported in literature mechanical impedance of human

elbow joint, (Zhang and Rymer, 1997). The parame-

ters obtainedin the experiments carried out are similar

to those values.

Correlating EMG–signals with the computed me-

chanical impedance can be considered as a function

of EMG-activity, according to Equation 3. Currently,

this functional relation has being found out.

4 CONCLUSIONS AND FUTURE

WORKS

NeuroLab is based on an upper limb robotic exoskele-

ton with which speciﬁc force proﬁles can be applied.

It establish a real multimodal interaction between the

user and the powered exoskeleton through a set of

BIODEVICES 2008 - International Conference on Biomedical Electronics and Devices

0 1 2 3 4 5 6

0.2

0.4

Estimated Parameters of Mechanical Impedance

Mean and standard deviation

Inertia

[N.m.s

/rad]

0 1 2 3 4 5 6

Viscosity

[N.m.s/rad]

0 1 2 3 4 5 6

Trial

Stiffness

[N.m/rad]

Figure 4: Mean and standard deviation of estimated param-

eters, for inertia (top), viscosity (middle) and stiffness (bot-

tom).

smart devices. With the networked platform, seve-

ral different experiments can be conﬁgured to explore

the human neuromotorsystem and to study the human

movement.

In the platform, there are independent devices that

communicate with each other, based on a Personal

Area Network (PAN) concept. Each device has a spe-

ciﬁc function and helps to address the overall goal of

the platform.

The system can be used in a wide range of ap-

plications. The results obtained with NeuroLab pro-

vide valuable information for robotics, modelling of

the human motor system, rehabilitation programs in

health care, training programs and biomechanics.

Lately, several studies are being conducted with

NeuroLab. The experiments presented in the paper

aim to estimate the properties of the human elbow

joint impedance and to obtain the viscoelasticity–

EMG relationships. System identiﬁcation is achieved

by perturbation analysis, using an external perturba-

tion application that produces changes in the dynam-

ics of system and EMG patterns.

The presented method to estimate the mechanical

impedance of the human arm is suitable to be used

in a clinical setting, e.g., with people with stroke un-

dergoing robotic rehabilitation for a paralyzed arm,

(Palazzolo et al., 2007).

Future work includes a quantitative analysis, pro-

cessing and correlation of the acquired signals (bio-

electric and biomechanical signals), based in Equa-

tion 3. Currently the EEG Monitoring Module is be-

ing validated and integrated in the system presented.

REFERENCES

Clancy, E. A. and Hogan, N. (1997). Relating agonist-

antagonist electromyograms to joint torque dur-

ing isometric, quasi-isotonic, nonfatiguing contrac-

tions. IEEE Transactions On Biomedical Engineering,

44(10):1024–1028.

DeLuca, C. J. (1997). The use of surface electromyography

in biomechanics. Journal of Applied Biomechanics,

13(2):135–163.

Dolan, J. M., Friedman, M. B., and Nagurka, M. L. (1993).

Dynamic and loaded impedance components in the

maintenance of human arm posture. IEEE Transac-

tions on Systems, Man and Cybernetics, 23(3):698–

709.

Hogan, N. (1984). Adaptive control of mechanical

impedance by coactivation of antagonist muscles.

IEEE Transactions On Automatic Control, 29(8):681–

690.

Kearney, R. E. and Hunter, I. W. (1990). System identiﬁ-

cation of human joint dynamics. Critical Reviews on

Biomedical Engineering, 18:55–87.

Krebs, H. I., Hogan, N., Aisen, M. L., and Volpe, B. T.

(1998). Robot-aided neurorehabilitation. IEEE Trans-

action On Rehabilitation Engineering, 6(1):75–87.

Palazzolo, J. J., Ferraro, M., Krebs, H. I., , Lynch, D., Volpe,

B. T., and Hogan, N. (2007). Stochastic estimation

of arm mechanical impedance during robotic stroke

rehabilitation. IEEE Transaction On Neural Systems

and Rehabilitation Engineering, 15(1):94–103.

Pfurtscheller, G., Muller, G., and Korisek, G. (2002). Men-

tal activity hand orthosis control using the eeg: a case

study. Rehabilitation, 41(1):48–52.

Rocon, E., Belda-Lois, J. M., Ruiz, A. F., Manto, M., and

Pons, J. L. (2007). Design and validation of a rehabil-

itation robotic exoskeleton for tremor assessment and

suppression. IEEE Transactions on Neural Systems

and Rehabilitation Engineering, 15(3):367–378.

Rosen, J., Brand, M., Fuchs, M., and Arcan, M. (2001). A

myosignal-based powered exoskeleton system. IEEE

Transaction On Systems, Man and Cybernetics - Part

A: Systems and Humans, 31(3):210–222.

Ruiz, A. F., Forner-Cordero, A., Rocon, E., and Pons, J. L.

(2006). Exoskeletons for rehabilitation and motor

control. In Proceedings of the IEEE International

Conference on Biomedical Robotics and Biomecha-

tronics (BioRob).

Tsuji, T., Morasso, P. G., and Ito, K. (1995). Human hand

impedance characteristics during maintained posture.

Byological Cybernetics, 74(1):475–485.

Wolpaw, J. R., Birbaumer, N., McFarland, D. J.,

Pfurtscheller, G., and Vaughan, T. M. (2002). Brain-

computer interfaces for communication and control.

Clinical Neurophysiology, 113:767–791.

Zhang, L. Q. and Rymer, W. Z. (1997). Simultaneous and

nonlinear identiﬁcation of mechanical and reﬂex prop-

erties of human elbow joint muscles. IEEE Transac-

tions On Biomedical Engineering, 44(12):1192–1209.

NEUROLAB: A MULTIMODAL NETWORKED EXOSKELETON FOR NEUROMOTOR AND BIOMECHANICAL

RESEARCH