SYNCHRONIZATION OF ARM AND HAND ASSISTIVE

ROBOTIC DEVICES TO IMPART ACTIVITIES OF DAILY

LIVING TASKS

Duygun Erol

and Nilanjan Sarkar

Department of Electrical & Electronics Engineering, Yeditepe University, Kayisdagi, Istanbul, Turkey

Department of Mechanical Engineering, Vanderbilt University, Nashville, TN, U.S.A.

Keywords: Robot-assisted rehabilitation, robot-assisted rehabilitation for activities of daily living tasks, coordination of

arm and hand assistive devices, hybrid system model.

Abstract: Recent research in rehabilitation indicates that tasks that focus on activities of daily living (ADL) is likely to

show significant increase in motor recovery after stroke. Most ADL tasks require patients to coordinate their

arm and hand movements to complete ADL tasks. This paper presents a new control approach for robot

assisted rehabilitation of stroke patients that enables them to perform ADL tasks by providing controlled

and coordinated assistance to both arm and hand movements. The control architecture uses hybrid system

modelling technique which consists of a high-level controller for decision-making and two low-level

assistive controllers (arm and hand controllers) for arm and hand motion assistance. The presented

controller is implemented on a test-bed and the results of this implementation are presented to demonstrate

the feasibility of the proposed control architecture.

1 INTRODUCTION

Stroke is leading cause a disability that results in

high costs to the individual and society (Matchar,

1994). Literature supports the idea of using intense

and task oriented stroke rehabilitation (Cauraugh,

2005) and creating highly functional and task-

oriented practice environments (Wood, 2003) that

increase task engagement to promote motor

learning, cerebral reorganization and recovery after

stroke. The task-oriented approaches assume that

control of movement is organized around goal-

directed functional tasks and demonstrated

promising results in producing a large transfer of

increased limb use to the activities of daily living

(ADL) (Ada, 1994). The availability of such training

techniques, however, is limited by the amount of

costly therapist’s time they involve and the ability of

the therapist to provide controlled, quantifiable and

repeatable assistance to movement. Consequently,

robot-assisted rehabilitation that can quantitatively

monitor and adapt to patient progress, and ensure

consistency during rehabilitation has become an

active research area to provide a solution to these

problems.

MIT-Manus (Krebs, 2004), MIME (Lum,

2006), ARM-Guide (Kahn, 2006) and the

GENTLE/s (Loureiro, 2003) are the devices

developed for arm rehabilitation, whereas Rutgers

Master II-ND (Jack, 2001), the CyberGrasp

(Immersion Corporation), a pneumatically

controlled glove (Kline, 2005) and HWARD

(Takahashi, 2005) are used for hand rehabilitation.

Even though existing arm and hand

rehabilitation systems have shown promise of

clinical utility, they are limited by their inability to

simultaneously assist both arm and hand

movements. This limitation is critical because the

stroke therapy literature supports the idea that the

ADL-focused tasks (emphasis on task-oriented

training), which engage patients to perform the tasks

in enriched environments have shown significant

increase in the motor recovery after stroke. Robots

that cannot simultaneously assist both arm and hand

movements are of limited value in the ADL-focused

task-oriented therapy approach. It is possible to

integrate an arm assistive device and a hand assistive

device to provide the necessary motion for ADL-

focused task-oriented therapy. However, none of the

existing controllers used for robot-assisted

rehabilitation can be directly used for this purpose

because they are not suited for controlling multiple

systems in a coordinated manner. In this work, we

address the controller design issue of a robot-

assisted rehabilitation system that can

simultaneously coordinate both arm and hand

Erol D. and Sarkar N. (2008).

SYNCHRONIZATION OF ARM AND HAND ASSISTIVE ROBOTIC DEVICES TO IMPART ACTIVITIES OF DAILY LIVING TASKS.

In Proceedings of the Fifth International Conference on Informatics in Control, Automation and Robotics - ICSO, pages 5-12

DOI: 10.5220/0001476700050012

 SciTePress

motion to perform ADL tasks using an intelligent

control architecture. The proposed control

architecture uses hybrid system modeling technique

that consists of a high-level controller and two low-

level device controllers (e.g., arm and hand

controllers). The versatility of the proposed control

architecture is demonstrated on a test-bed consisting

of an upper arm assistive device and a hand assistive

device. Note that the presented control architecture

is not specific to a given arm and hand assistive

device but can be integrated with other previously

proposed assistive systems.

In this paper, we first present the intelligent

control architecture in Section 2, and then the

rehabilitation robotic system and design details of

the high-level controller are presented in Section 3.

Later, results of the experiments to demonstrate the

efficacy of the proposed controller architecture are

presented in Section 4. Finally, the contributions of

the work are presented in Section 5.

2 INTELLIGENT CONTROL

ARCHITECTURE

Let us first present the proposed intelligent control

architecture in the context of generic ADL tasks that

require coordination of both arm and hand

movement (e.g., eating, drinking, etc.). Stroke

patients may not be able to complete the ADL tasks

by themselves because of motor impairments. Thus,

low-level arm assistive controller and low-level

hand assistive controller may be used to provide

assistance to the subject’s arm and hand movement,

respectively. The nature of assistance given to the

patients and coordination of the assistive devices,

however, could be impacted by various events

during the ADL task (e.g., completion of a subtask,

safety related events etc.). A high-level controller

(HLC) may be used to allocate task responsibility

between the low-level assistive controllers (LLACs)

based on the task requirements and specific events

that may arise during the task performance. HLC

plays the role of a human supervisor (therapist) who

would otherwise monitor the task, assess whether

the task needs to be updated and determine the

activation of the assistive devices. However, in

general, the HLC and the LLACs may not

communicate directly because each may operate in

different domains. While the LLACs may operate in

a continuous way, the HLC may need to make

intermittent decisions in a discrete manner. Hybrid

system theory provides mathematical tools that can

accommodate both continuous and discrete systems

in a unified manner. Thus, we take advantage of

using a hybrid system model to design the proposed

intelligent control architecture (Koutsoukos, 2000).

In this architecture, the “Plant” represents both the

assistive devices and their low-level assistive

controllers and the Interface functions as analog-to-

digital/digital-to-analog (AD/DA) adaptor.

The proposed control architecture for robot-

assisted rehabilitation to be used to perform ADL

tasks is presented in Fig. 1. In this architecture, the

sensory information from the arm assistive device,

the hand assistive device and the feedback from the

human are monitored by the process-monitoring

module through the interface. The sensory

information (plant event) is converted to a plant

symbol so that the HLC can recognize the event.

Based on a plant symbol, the decision making

module of the HLC sends its decision to the LLACs

through the interface using the control symbols.

Interface converts the control symbols to the plant

inputs which are used to activate/deactivate the

LLACs to complete the ADL task. The proposed

control architecture is extendible in the sense that

new events can be included by simply monitoring

the new sensory information from the human and the

assistive devices, and accommodated by introducing

new decision rules.

Figure 1: Control Architecture.

3 METHODOLOGY

The primary focus of this paper is to demonstrate

how the assistive devices can be coordinated using

the proposed intelligent control framework for a

given ADL task. The intelligent control framework

consists of a HLC and two low-level arm and hand

assistive controllers. The focus of the paper is to

design of HLC that can coordinate a number of

given LLACs using the presented intelligent control

ICINCO 2008 - International Conference on Informatics in Control, Automation and Robotics

framework. First we briefly present the rehabilitation

system used as the test-bed to implement the

intelligent control framework. We then present a

detailed discussion on the design and

implementation of the HLC and its workings within

the presented control framework using an ADL task.

3.1 Rehabilitation Robotic System -

a Test Bed

The rehabilitation robotic system used in this work

consists of an arm assistive device, hand assistive

device and two sensory systems (contact detection

and proximity detection systems) (Fig. 2A). We

have modified a Power Grip Assisted Grasp Wrist-

Hand Orthosis (Broadened Horizons) as a hand

assistive device (Fig. 2B). A computer control

capability in a Matlab/Realtime Windows Target

environment is added in order to integrate the hand

device in the proposed control architecture. The

subject is asked to follow the opening/closing speed

of the hand device and if the subject cannot follow

the hand device movement, then the hand device

provides assistance to complement subject’s effort to

open/close his/her hand. The PUMA 560 robotic

manipulator is augmented with a force-torque sensor

and a hand attachment device (Fig.2A) to provide

assistance to the upper arm movement (Erol, 2007).

A proportional-integral-derivative (PID) position

control is used as a low-level arm assistive controller

for providing robotic assistance to a subject. The

subject is asked to pay attention to tracking the

desired position trajectory (visually monitoring

his/her actual and desired position trajectories on a

computer screen) as accurately as possible. If the

subject deviates from the desired motion, then the

low-level arm assistive controller provides robotic

assistance to complement the subject’s effort to

complete the task as required. We have designed a

contact detection system to provide sensory

information about grasping activity that may be a

part of an ADL task of interest. The force-sensitive

resistors (FSR) (Interlink Electronics, Inc.) are

placed on the fingertip to estimate the forces applied

on the object during the grasping task (Fig. 2C).

When the subject starts grasping an object, then the

voltage across the FSR changes as a function of the

applied force. Additionally, a proximity detection

system (PDS) is designed in order to detect the

closeness of the subject’s hand relative to the object

to be grasped. The PDS contains a phototransistor

(sensitive to infrared light) and an infrared emitter,

which are mounted onto two slender posts close to

the object facing each other. When the subject

approaches to the object by moving his/her hand

between these posts, the continuity of the receiving

signal (infrared beam) is broken and the

corresponding voltage change is used to generate an

event to inform the HLC that the subject is close to

the object.

Figure 2A: Rehabilitation Robotic System, Figure 2B:

Hand Assistive Device and Figure 2C: Force Sensor

Resistors Placement on the Fingers.

3.2 Task Description

The main focus of this work is to present how a

HLC is designed and how it functions within the

proposed intelligent control architecture. The ADL

tasks consist of several primitive movements such as

reaching towards an object, grasping the object,

lifting the object from the table, using the object for

eating/drinking, and placing the object back on the

table (Murphy, 2006) and they all require

coordination between arm and hand movements. In

here, we choose one of the ADL tasks, called

drinking from a bottle (DFB) task to explain the

HLC. We decompose the DFB task into the

following primitive movements: i) reach towards the

bottle while opening the hand, ii) reach the bottle,

iii) close the hand to grasp the bottle, iv) move the

bottle towards the mouth, v) drink from a bottle

using a straw, vi) place the bottle back on the table,

vii) open the hand to leave the bottle back on the

table and viii) go back to the starting position. Note

that similar task decomposition could be defined for

other ADL tasks.

3.3 Design Details of the High-Level

Controller

3.3.1 Theory of Hybrid Control Systems

The hybrid control systems consist of a plant which

is generally a continuous system to be controlled by

a discrete event controller (DES) connected to the

plant via an interface in a feedback configuration

(Koutsoukos, 2000). If the plant is taken together

with the interface, then it is called a DES plant

SYNCHRONIZATION OF ARM AND HAND ASSISTIVE ROBOTIC DEVICES TO IMPART ACTIVITIES OF DAILY

LIVING TASKS

model. The DES plant model is is a nondeterministic

automaton which is represented by

G ( P,X ,R, , )



. Here



is the set of discrete

states;



is the set of plant symbols generated

based on the events; and



is the set of control

symbols.

:P R 2

×→





is the state transition

function. For a given DES plant state and a given

control symbol, state transition function is defined as

the mapping from

R×



to the power set



, since

for a given plant state and a control symbol the next

state is not uniquely defined. The output

function,

:P P 2

×→





, maps the previous and

current plant states to a set of plant symbols. The

DES controller, which is called the HLC in this

work, controls the DES plant. The HLC is

responsible to coordinate the assistive devices based

on both task and the safety requirements. The HLC

is modeled as a discrete-event system (DES)

deterministic finite automaton, which is specified

(S,X,R, , )





. Here



is the set of control

states. Each event is converted to a plant symbol,

where



is the set of such symbols, for all discrete

states. The next discrete state is activated based on

the current discrete state and the associated plant

symbol using the following transition

function:

:S X S

×→





. In order to notify the LLACs

the next course of action in the new discrete state,

the HLC generates a set of symbols, called control

symbols, denoted by



, using an output

function:

:S R

→



3.3.2 Modelling of an ADL Task using

Hybrid Control System

Now we discuss how the above theory could be used

to model and control an ADL task (e.g., the DFB

task) for rehabilitation therapy. The first step is to

design the DES plant and define the hypersurfaces

that separates different discrete states. The

hypersurfaces are used to detect the events and are

decided considering the capabilities of the

rehabilitation robotic systems and the requirements

of the task. The following hypersurfaces are defined:

hvir0,h xx

−∈=> =≥

(

)

(

)

h vfsr vth hcb 0

=<∧==

()

hxx (hob0),

−∈=≤ ∧ ==

(

)

htt

hand

===

θθ

=<<

rth

ττ

=≥

hhth

ττ

=≥

()

heb1

===

(

)

hpb1

===

(

)

(

)

hpb0eb0

===∧==

, where vir is the voltage in

the PDS system.

and

are the hand actual

position and the object’s position, respectively.

∈

a value used to determine if the subject is close

enough to the object’s position.

vfsr and vth are the

voltage across the FSRs and the threshold voltage,

respectively. The values of

hob and hcb are binary

values, which could be 1 when it is pressed and 0

when it is released.

t and

hand

are the current time

and the final time to complete hand opening,

respectively.

and

represent the set of lower and

upper limits of the joint angles, respectively and

the set of the actual joint angles.

and

rth

are the

torque applied to the motor of the arm assistive

device and the torque threshold value, respectively.

The torque applied to the motor of the hand assistive

device and its threshold value is defined as

and

hth

, respectively. The values of eb and

b are

binary values, which could be 1 when it is pressed

and 0 when it is released. The above hypersurfaces

can be classified into two groups: i) the

hypersurfaces that are defined considering the

requirements of the tasks (i.e.,

), and ii) the

hypersurfaces that are defined considering the

capabilities of the rehabilitation robotic system (i.e.,

). The hypersurfaces provide information to

the HLC in order to make decisions for execution of

the task in a safe manner. The set of DES plant

states



is based upon the set of hypersurfaces

realized in the interface. Each region in the state

space of the plant, bounded by the hypersurfaces, is

associated with a state of the DES plant. During the

execution of the task, the state evolves over time and

the state trajectory enters a different region of the

state space by crossing the hypersurfaces and a plant

event, occurs when a hypersurface is crossed. A

plant event generates a plant symbol to be used by

the HLC. The plant symbols



in the DES plant

model

G ( P, X ,R, , )



are defined as follows:

[n] ( p[n 1],p[n])

−



(1)

where

X∈



∈



is the output function and n

is the time index that specifies the order of the

symbols in the sequence. In (1) the plant symbol,



is generated as an output function of the current and

the previous plant state. We define the following

plant symbols considering the hypersurfaces

discussed before: i)



, the arm approaches to the

bottle with the desired grip aperture, which is

generated when

is crossed, ii)



, the arm reaches

to the bottle, which is generated when

is crossed,

iii)



, the hand reaches desired grip closure to

ICINCO 2008 - International Conference on Informatics in Control, Automation and Robotics

grasp the bottle, which is generated when

crossed, iv)



, the arm leaves the bottle on the table,

which is generated when

is crossed, v)



, the

hand reaches desired grip aperture, which is

generated when

is crossed, vi)



, safety related

issues happened such as the robot joint angles are

out of limits (when

is crossed), or the robot

applied torque is above its threshold (when

crossed), or hand device applied torque is above its

threshold (when

is crossed) or emergency button

is pressed (when

is crossed), vii)



, the subject

presses the pause button, which is generated when

is crossed, and viii)



, the subject releases the

pause button which is generated when

is crossed.



{

}

,x ,x ,x ,x ,x ,x ,x

45 7

123 6 8

    

is the set of plant

symbols.

In this work, the purpose of the DES controller

(HLC) is to activate/deactivate the assistive devices

in a coordinated manner to complete the DFB task.

In order to perform this coordination, the following

control states are defined:



: both the hand device

and arm device are active, ii)



: the arm device

alone is active, iii)



: the hand device alone is

active to close the hand, iv)



: the hand device

alone is active to open the hand, v)



: both the arm

and hand devices are idle. Additionally, a memory

control state (



) is defined to detect the previous

control actions when the subject wants to continue

with the task after he/she presses pause button.

{

}

S s ,s ,s ,s ,s ,s

123456



   

is the set of control states in

this application. When new control actions are

required for an ADL task, new control states can be

included in the set of the states,



. The transition

function

:S X S

×→





uses the current control state

and the plant symbol to determine the next control

action that is required to update the ADL task, where

S∈



X∈



rR∈



, and n is the time index that

specifies the order of the symbols in the sequence.

[n] (s[n 1],x[n])

=−



(2)

The HLC generates a control symbol



, which is

unique for each state,



as given in (3). In here, the

following control symbols are defined: i)



: drive

arm device towards the object while driving hand

device to open the hand, ii)



: drive arm device to

perform various primitive arm motion such as move

the bottle towards the mouth etc., iii)



: drive hand

device to close the hand to grasp the bottle, iv)



drive hand device to open the hand to leave the

bottle, and v)



: make arm and hand devices idle.

The set of control symbols are defined as

{

}

R r ,r ,r ,r ,r

12345



   

r[n] (s[n])



(3)

The LLACs cannot interpret the control symbols

directly. Thus the interface converts the control

symbols into continuous outputs, which are called

plant inputs. The plant inputs are then sent to the

LLACs to modify the ADL task. We define the

following plant inputs: i) if



then provide 1 to

activate both arm and hand devices, ii) if



, then

provide 2 to activate only the arm device, iii) if



, then provide 3 to activate only the hand

device to close hand, iv) if



, then provide 4 to

activate only the hand device to open hand, v) if



, then provide 0 to keep both arm and hand

devices idle. Note that the design of the elements of

the DES plant and the DES controller is not unique

and is dependent on the task and the sensory

information available from the robotic system.

4 RESULTS

In this section we present two experiments that were

conducted to demonstrate the feasibility and

usefulness of the proposed control architecture.

Since we experiment with an unimpaired subject

who could ideally do the DFB task by himself

(unlike a real stroke patient) we instructed him to be

passive so that we can demonstrate that the proposed

control architecture was solely responsible for the

coordinated arm and hand movements (which is the

main objective of this work) as needed to complete

the DFB task. Such an experimental condition is not

only helpful to demonstrate the efficacy of the

proposed control architecture but also could occur

when a low functioning stroke survivor participates

in a task-oriented therapy who will initially need

continuous robotic assistance to perform an ADL

The subject was asked to wear the hand device

and then place his forearm on the hand attachment

(Fig. 2A). In the first experiment (E1), we asked the

subject to perform the DFB task, where the task

proceeded as planned (i.e., there was no event

SYNCHRONIZATION OF ARM AND HAND ASSISTIVE ROBOTIC DEVICES TO IMPART ACTIVITIES OF DAILY

LIVING TASKS

occurred during the task that would require dynamic

modification of the execution of the task; however, it

still needed the necessary coordination between

hand and arm motion). We designed a DFB task

trajectory in consultation with a physical therapist as

shown in Fig. 3. Fig. 3 shows how the DFB task was

supposed to proceed: the subject was required to

reach the bottle (A-B), grasp the bottle by applying a

certain amount of force (B-C), bring the bottle to the

mouth (C-D), drink water (D-E), bring back the

bottle from where he picked it up (E-F), open hand

to release the bottle (F-G), and then go back to the

starting position (G-H). Furthermore, the desired

trajectory from A-B into A-A’ and A’-B trajectories

have been decomposed because in naturalistic

movement it has been shown that a subject reached

his/her maximum aperture approximately two-third

of the way through the duration of the reaching

movement (Jeannerod, 1981).

Figure 3: Desired Motion Trajectories for a DFB task.

The overall mechanism for the high-level control,

which is used to activate/deactivate the LLACs, is

shown in Fig. 4. When a device is active, we mean it

tracks a non-zero trajectory and when the device is

idle, we mean the device remains in its previous

position set points.

Let us now explain how the HLC accomplished

the DFB task using the control mechanism given in

Fig. 4. When the DFB task started,



became active

where both arm and hand devices remained active

till point A’ to help the subject to open his hand

while moving towards the bottle. Then at point A’,

the

vir crosses a predefined threshold value, which

confirmed that the subject reached close to the bottle

(



was generated) and



state became active. The

arm device remained active to help the subject to

reach the bottle and the hand device was idle from

A’ to B. After that at point B, when the subject’s

position,

, was close to the bottle position,

, and

then



was generated and



state became active. If



was active, then the hand device remained active

to assist the subject to grasp the bottle. Then, at

point C, the

vfsr value was dropped below the

threshold (



was generated) and



state became

active again. The arm device remained active to

assist the subject to move the bottle to his mouth, to

drink water using a straw and at the end to leave the

bottle on the table. When the subject brought the

bottle back on the table at point F,



was generated

and



state became active and the hand device

remained active to help the subject to open his hand

till G. Then the subject reached the desired grip

aperture

(

)

hand



was generated and



state

became active. The arm device remained active to

help the subject to go back to the starting position.

The actual trajectory of the subject was exactly same

as the desired trajectory given in Fig. 3. The

subject’s hand configuration diagram was given in

Fig. 5. It could be seen from the figures that the

subject was able to track the desired trajectories

while opening/closing his hand at desired times.

Figure 4: Control Mechanism for the HLC.

Figure 5: Hand Configuration Diagram for E1.

In the second experiment (E2), we demonstrated that

if an event takes place at some point of time during

the task execution that requires modification of the

desired task trajectory such as a stroke patient wants

to pause for a while due to some discomfort, then the

HLC has the ability to dynamically modify the

desired trajectory using the control mechanism given

in Fig. 4. In this case, the subject started the

execution of the task with the same desired

trajectory as shown in Fig. 3 (which is the dotted

trajectory in Fig. 6). During the execution of the task

ICINCO 2008 - International Conference on Informatics in Control, Automation and Robotics

at time t’, the subject pressed the pause button at

time t’ (when



was active), the plant symbol



was generated and



state became active and both

the arm and hand devices became idle.



state

stored the previous active state. When the subject

released the pause button at time

tt’ to continue the

task execution,



was generated and



became

active again to activate both the arm and hand

devices to resume the task execution. The rest of the

desired trajectory had been generated in the same

way as it was described in E1 (Fig. 6-solid lines). It

could be noticed from Fig. 6-solid lines that at time

t’, the assistive devices remained in their previous

set points. Additionally, the subject’s position at

time of the

tt’ was automatically detected and taken

as an initial position to continue the task where it

was resumed with zero initial velocity (Fig. 6-solid

lines). If the HLC did not modify the desired

trajectories to register the intention of the subject to

pause the task, then i) the desired motion trajectories

would start at point

tt’ with a different starting

position and a non-zero velocity (Fig. 6 -dotted

lines), which could create unsafe operating

conditions, and ii) the subject would close/open his

hand at undesirable times. We had also noticed that

the subject’s actual trajectory was same as given in

Fig. 6-solid line. Fig. 7 demonstrated the subject’s

hand configuration diagram for E2. It can be noticed

that the subject was able to track the modified

desired trajectory and he was able to coordinate his

arm and hand motions in a safe and desired manner.

Figure 6: Desired Trajectory for the DFB task when an

Unplanned Event Happened.

It is conceivable that one could pre-program all

types of desired trajectories beforehand such that

they could address all types of unplanned events,

and retrieve them as needed. However, for non-

trivial tasks, designing such a mechanism might be

too difficult to manage and extend as needed

Figure 7: Hand Configuration Diagram for E2.

5 DISCUSSION/ CONCLUSIONS

The purpose of this work is to design a versatile

control mechanism to enable robot-assisted

rehabilitation in a task-oriented therapy that involves

tasks requiring coordination of motion between arm

and hand. In order to achieve this goal we design a

new intelligent control architecture that is capable of

coordinating both the arm and the hand assistive

devices in a systematic manner. This control

architecture exploits hybrid system modeling

technique to provide the robotic assistance to enable

a subject to perform ADL tasks that may be needed

in a task-oriented therapy and has not been explored

before for rehabilitation purpose. Hybrid system

modeling technique offers systematic control design

tools that provide design flexibility and extendibility

of the controller, which gives ability to integrate

multiple assistive devices and to add/modify various

ADL task requirements in the intelligent control

architecture. The control architecture combines a

high-level controller and low-level assistive

controllers (arm and hand). In here, the high-level

controller is designed to coordinate with the low-

level assistive controllers to improve the robotic

assistance with the following objectives: 1) to

supervise the assistive devices to produce necessary

coordinated motion to complete a given ADL task,

and 2) to monitor the progress and the safety of the

ADL task such that necessary dynamic

modifications of the task execution can be made to

complete the given task in a safe manner.

Although the focus of the current work is to

present a new high-level control methodology for

rehabilitation that is independent of the low-level

controllers, we want to mention the limitations of the

hand and arm assistive devices used in the presented

work. The hand device used in this paper does not

allow independent control of fingers in performing

various hand rehabilitation tasks. As discussed

earlier, the focus of the paper is to present how arm

and hand motion can be dynamically coordinated to

accomplish ADL tasks. In that respect, the current

hand device allowed us to perform an ADL task that

showed the efficacy of the presented high-level

controller. A more functional hand device would

SYNCHRONIZATION OF ARM AND HAND ASSISTIVE ROBOTIC DEVICES TO IMPART ACTIVITIES OF DAILY

LIVING TASKS

allow the patients to perform more complex ADL

tasks. Note that this current hand device is being

used with C5 quadriplegic patients to complete their

ADL tasks such as picking up bottle etc. (Broadened

Horizons). We are also aware that a PUMA 560

robotic manipulator might not be ideal for

rehabilitation applications. However the use of

safety mechanisms, both in hardware (e.g.,

emergency button, quick arm release mechanism

etc.) and in software (e.g., within the design of the

high-level controller) will minimize the scope of

injuries. Note that the proposed control architecture

is not specific to the presented assistive devices but

can also be integrated with other assistive devices.

We believe that such a robot-assisted

rehabilitation system with capabilities of

coordination of both arm and hand movement is

likely to combine the advantages of robot-assisted

rehabilitation systems with the task-oriented therapy.

In this paper, the efficacy of the proposed intelligent

controller is demonstrated with healthy human

subject. We are aware that a stroke patient with a

spastic arm is much more different from a healthy

subject following the robotic moves. In that respect,

more functional assistive devices and their

corresponding low-level controllers can be

integrated inside the proposed intelligent controller

to allow stroke patients to take part in task-oriented

therapy. As a future work, it is possible to use

intelligent robot-assisted rehabilitation systems in

clinical trials to understand on how impairment

changes carryover of gained functional abilities to

real living environments and how robot-assisted

environments influence these changes.

ACKNOWLEDGEMENTS

We gratefully acknowledge the help of Dr. Thomas

E. Groomes and Sheila Davy of Vanderbilt

University's Stallworth Rehabilitation Hospital for

their feedback about task design and Mark Felling

who is C5 quadriplegic patient for his feedback

about the hand assistive device.

REFERENCES

Ada, L., Canning, C. G., Carr, J. H., Kilbreath, S. L. &

Shepherd, R. B. (1994). Task-specific training of

reaching and manipulation. In Insights into Reach to

Grasp movement, 105, 239-265.

Broadened Horizons, http://www.broadenedhorizons.com/

Cauraugh J. H., Summers, J.J. (2005). Neural plasticity

and bilateral movements: A rehabilitation approach for

chronic stroke. Prog Neurobiol., 75, 309-320.

Erol, D. & Sarkar, N. (2007), Design and Implementation

of an Assistive Controller for Rehabilitation Robotic

Systems. Inter. J. of Adv. Rob. Sys, 4(3), 271-278.

Immersion Corporation, http://www.immersion.com/.

Jack, D., Boian, R., Merians, A. S., Tremaine, M., Burdea,

G. C., Adamovich, S. V., Recce, M. & Poizner, H.

(2001). Virtual reality-enhanced stroke rehabilitation.

IEEE Transactions on Neural Systems and

Rehabilitation Engineering, 9, 308–318.

Jeannerod, M., (1981). ‘Intersegmental coordination

during reaching at natural visual objects’. In

Attention and performance IX, J. L. a. A. Baddeley,

Ed. New Jersey, (pp. 153-168).

Kahn, L. E., Zygman, M. L., Rymer, W. Z. &

Reinkensmeyer, D. J. (2006). Robot-assisted reaching

exercise promotes arm movement recovery in chronic

hemiparetic stroke: a randomized controlled pilot

study. J. Neuroengineering Rehabil, 3,1-13.

Kline, T., Kamper, D. & Schmit, B. (2005). Control

system for pneumatically controlled glove to assist in

grasp activities. IEEE 9th International Conference on

Rehabilitation Robotics, pp. 78 - 81.

Krebs, H. I., Ferraro, M., Buerger, S. P., Newbery, M. J.,

Makiyama, A., Sandmann, M., Lynch, D., Volpe, B.

T. & Hogan, N. (2004). Rehabilitation robotics: pilot

trial of a spatial extension for MIT-Manus. J

Neuroengineering Rehabil, 1, 1-15.

Koutsoukos, X. D., Antsaklis, P. J., Stiver, J. A. &

Lemmon, M.D. (2000). Supervisory control of hybrid

systems. IEEE on Special Issue on Hybrid Systems:

Theory and Applications, 88, 1026-1049.

Matchar D. B., Duncan, P. W. (1994). Cost of Stroke,

Stroke Clin. Updates. 5, 9-12.

Murphy, M. A., Sunnerhagen, K. S., Johnels, B. & Willen,

C. (2006). Three-dimensional kinematic motion

analysis of a daily activity drinking from a glass: a

pilot study. J. Neuroengineering Rehabil., 16, 3-18.

Loureiro, R., Amirabdollahian, F., Topping, M., Driessen,

B. & Harwin, W. (2003). ‘Upper limb mediated stroke

therapy - GENTLE/s approach. Autonomous Robots,

15, 35-51.

Lum, P. S., Burgar, C. G., Van der Loos, H. F. M., Shor,

P. C., Majmundar, M. & Yap, R. (2006). MIME

robotic device for upper-limb neurorehabilitation in

subacute stroke subjects: A follow-up study. J. of

Rehab. Res. & Dev., 43, 631-642.

Takahashi, C. D., Der-Yeghiaian, L., Le, V. H. & Cramer,

S. C. (2005). A robotic device for hand motor therapy

after stroke. IEEE 9th International Conference on

Rehabilitation Robotics, pp. 17 – 20.

Wood, S. R., Murillo, N., Bach-y-Rita, P., Leder, R. S.,

Marks, J. T. & Page, S. J. (2003).Motivating, game-

based stroke rehabilitation: a brief report. Top Stroke

Rehabil, 10, 134-140.

ICINCO 2008 - International Conference on Informatics in Control, Automation and Robotics