PRACTICAL DESIGN OF FULL BODY EXOSKELETONS

Stretching the Limits of Weight and Power

Stefan Roland Taal and Yoshiyuki Sankai

Graduate School of Systems and Information Engineering, University of Tsukuba

1-1-1, Tennodai, Tsukuba, Ibaraki, 305-8573, Japan

Keywords: Robot suit, HAL, Exoskeleton, Augmentation, Practical design, Exo-spine, Nurses, Lifting assist.

Abstract: The development of full body, wearable exoskeletons has been limited by the constraints of weight and

available power. Because of this it has not been possible to create one that augments all DoF of its human

wearer with enough power to assist, e.g., nurses and other workers. To achieve more usefulness despite the

limitations, a practical design approach that considers the motions and needs of the wearer is an appropriate

tool to reveal new opportunities. This approach was used to find solutions for a fully supported 3DoF exo-

spine, supported shoulder girdle motion, and other challenges that have so far received little or no attention.

No extra actuators are required, thus adding a minimum to weight and power. The improvements found

using this practical approach suggest related fields like rehabilitation could profit as well.

1 INTRODUCTION

Recent levels of technology have enabled the

creation of various exoskeletal devices: robots that

surround (parts of) a human wearer in order to assist

him in his movements. Applications range from

rehabilitation to strengthening nurses and others in

their work. Yet, the all-round applicability of fully

wearable, i.e. also energetically autonomous, and

particularly full body exoskeletons, has so far been

limited by the low amount of degrees of freedom

(DoFs) and actuators achievable in such devices.

Increasing the applicability requires augmen-

tation of more human DoFs. This necessitates

adding heavy actuators and an accordingly larger

power supply for a running time of perhaps a few

hours. The more useful and thus larger the device

the more unlikely it is to fit in the settings of a

hospital or home, and hence designers were forced

to limit the abilities of their exoskeletons.

Considering the needs of aging societies to take

care of the older generations, this research focuses

on exoskeletons for augmentation of nurses and

other workers, and has a long term goal to develop a

version for physically challenged patients. It is based

on the full body robot suit HAL-5 from which a

lower body suit was derived for patients who have

difficulties walking (Suzuki et al. 2005).

To arrive at new solutions, this paper reviews

existing exoskeletons, shows why and how we

should change our design approach, and, to show the

effectiveness of the new approach, proposes

solutions from a mechanical perspective that

maximize the capabilities of full body robot suits.

1.1 Existing Exoskeletons

Lower body exoskeletons have been discussed

(Dollar & Herr 2008) and few challenges remain. As

for the upper body part (from the hip to the hands)

(a) (b)

Figure 1: Lifting DoF: the interdependence between hip en

trunk moments during lifting, as indicated by the arrows in

(a). The interaction forces, (b), between different body

parts provide additional proof.

133

Roland Taal S. and Sankai Y. (2010).

PRACTICAL DESIGN OF FULL BODY EXOSKELETONS - Stretching the Limits of Weight and Power.

In Proceedings of the Third International Conference on Biomedical Electronics and Devices, pages 133-138

DOI: 10.5220/0002756901330138

 SciTePress

there are more DoFs and larger workspaces that

ultimately compete with the constraints of weight

and power. Hence, this is the focus of this review.

One wearable full body exoskeleton is the nurse

power suit (Yamamoto 2002). It uses a pneumatic

actuator system to augment the muscles used for

lifting patients. While it is focused on patient lifting,

its workspace, however, is otherwise limited.

Another, the Agri Robot, has not yet appeared in

print, but may be found on the web (Toyama 2009).

It actuates motors that coincide with the knees, hips,

shoulders and elbows according to spoken

commands. Its main purpose is helping farmers.

These two exoskeletons, as well as HAL, show

exactly how the limitations on weight and power

result in augmentation of few DoFs while the

shoulder girdle and spine remain immobile.

As for other types, there are several wearable

arm exoskeletons that augment all DoFs of the

human arms and shoulder girdle (Schiele & Van der

Helm 2006) (Folgheraiter et al. 2009). These are for

rehabilitation and haptics and require only small

output torques. Using such structures to assist lifting,

however, would result in larger and heavier devices.

The XOS exoskeleton, manufactured by Sarcos,

also remains unpublished (BBC News 2008). This

full body suit requires an external power source, but

can provide powerful augmentation. The robot’s

arms only interact with the human at the end effector,

thereby allowing the shoulder girdle to move as well.

Another type of exoskeletal devices consists of

arms supported on a fixed base. The purpose of such

devices differs, but, despite the freedom regarding

weight and power, girdle motion has received limi-

ted attention (Perry & Rosen 2006) (Liszka 2006).

Lastly, pneumatic muscle actuators have been

used in a full body (Tsagarakis & Caldwell 2003)

and an upper body exoskeleton (Aida et al. 2009),

and have been shown to provide the torque required

for lifting. They work like muscles, making them

very compatible with humans. The main challenge,

however, is to make a wearable power supply.

So far it can be concluded that, using current

technology, being wearable and energetically

autonomous cannot be combined with having all

DoFs active and powerful enough to lift, e.g.,

patients. Critically, shoulder girdle motion has not

been implemented in a full body exoskeleton, and

spine motion has not received any attention at all.

1.2 Towards a New HAL

The current full body HAL suit, HAL-5, shown in

Fig. 1a, consists of frames interconnected by power

units, which each contain an electromotor and

reduction gears, positioned directly next to the hip,

knee, shoulder (flexion) and elbow joints of the

wearer to assist his movements. Additional passive

DoFs are located at each shoulder, upper arm, and

ankle joint. The suit is powered by batteries.

The system is controlled according to the

intentions of the wearer, which are obtained by

measuring the bioelectric signal (BES) on the skin

above the main flexor and extensor muscles

associated with each augmented human joint. Motor

torques are calculated according to these signals.

It is expected that similar control techniques and

actuators will be used in the new version. In addition,

the wearer is assumed to be a healthy person.

Considering the found limitations and the aim to

aid nurses, a new design approach for HAL should:

1) Achieve the most practicality given limited

technology;

2) Enable handling of patients by nurses, by

supporting the forces typically exerted by them.

The word ‘practicality’ in the first goal implies

“fitted for actual work or activities”, and is

considered the main property to ensures HAL’s

usability in our human society.

2 A DIFFERENT APPROACH

2.1 Challenges

These goals inevitably pose several specific

challenges. Firstly, not actuating some DoFs in order

to save weight and power poses the dilemma of

creating either passive DoF or a fixed structure

instead. Passive means that the wearer will

occasionally be required to exert a high degree of

effort to handle heavy objects, whereas fixing

reduces the human workspace and can result in high

forces between the wearer and the robot.

When considering the practical usage of an

exoskeleton it may be seen that both during daily

tasks (Rosen et al. 2005) and during working (Vieira

& Kumar 2004) gravity forces are the most

prevalent. Although several exoskeletons

specifically counter the forces of gravity during

lifting (Suzuki et al. 2005) (Yamamoto, 2002)

(Toyama, 2009), this focus also strongly limits the

workspace by limiting various DoFs. Moreover, the

loads should never be transferred from the suit to the

wearer. E.g., as will be shown in section 4, the load

supported by the suit may bear upon the wearer’s

body during walking. Some guarantee that the suit

BIODEVICES 2010 - International Conference on Biomedical Electronics and Devices

134

compensates gravity and transfers its weight and that

of the carried load directly to the floor is necessary.

Considering patient-handling by nurses it can be

seen that pushing and pulling forces are prevalent as

well, e.g., when turning a patient around in bed

(Schibye et al., 2003). A practical exoskeleton will

thus have to be able to support these forces as well.

Next, skin irritation around fastening equipment

is a problem not often considered during design, but

mostly revealed by experiments (Hidler & Wall

2005) (Colombo et al. 2000). Schiele and Van der

Helm (2006) showed how this partly arises from un-

avoidable misalignments between wearer and robot.

Regarding augmentation of the hands, which

would be necessary for picking up heavy objects, it

can be seen that only some fully actuated arms have

wrist actuators (Schiele & Van der Helm 2006)

(Folgheraiter et al. 2009) (Perry & Rosen 2006)

whereas for the fingers there are only rehabilitation

devices (Sasaki et al. 2004) (Mulas et al. 2005).

Unfortunately, all these devices also indicate that it

is very difficult to augment fingers up to a practical

load of around 25kg for one hand.

Lastly, implementing shoulder girdle and spine

movability requires two DoFs for each shoulder and

three for the spine, totalling seven extra actuators.

This would almost double the amount on HAL.

2.2 A Human Practical Approach

Exoskeletal structures are typically designed using a

machine approach, basing the design on the range of

motion (RoM) and torques of the human joints that

they interact with. For wearable robot suits it seems

that with this approach current challenges cannot be

overcome. On the other hand, the ways we use our

bodies for work reveal characteristics that may

provide unknown design opportunities.

In order to discover new solutions this paper

posits that, although the number of postures and

motions that may be achieved with the many DoFs

our bodies provide is very large, we only use a

limited subset of them in our daily lives and work

because they are somehow optimal. If a robot suit

can support this limited, practical set of postures and

motions, then it may be considered sufficient.

To illustrate, it is possible for people to eat while

maintaining their elbows at shoulder height. People

generally avoid this because it is tiring. It is not

practical. A practical design approach would

therefore consider what the wearer actually needs,

wants, and does when wearing an exoskeleton.

What the wearer primarily needs is gravity

compensation and the ability to move in ways that

tasks may be performed as desired, without feeling

the weight of the suit. Also, the suit must know the

wearer’s intentions, as was realized with HAL’s

intention based control.

In particular the motions that are desirable or

biomechanically optimal or motions otherwise used

in practice enable new solutions by requiring HAL

to assist only certain, instead of all possible activities.

E.g., the way an object is lifted reveals where and

when augmentation is required. This is discussed

further in the next section.

3 A SEMI-ACTIVE EXO-SPINE

3.1 Unified DoFs

Heavy objects, or patients, are generally too large to

hold on one side and are usually held in front of the

wearer. Additionally, holding heavy objects on the

side with one hand is unbalancing during walking

and is not doable beyond normal human strength

without sufficient hand augmentation, which, as

mentioned, does not exist.

When lifting objects in front the various muscles

activated in the hips and back compose several DoFs.

However, observing how they are activated, as

shown in Fig. 1a, it can be seen that in the hips and

back the moments are all generated in the same

direction. They act as a single unified DoF.

Some validation can be obtained from Fig. 1b.

By separating the trunk, pelvis, and legs the

interaction forces can be drawn schematically. This

shows that during lifting - knowing that no other

external forces are applied to the pelvis - the

direction of the moments in the hips must always be

the same as throughout the spine. Additionally, this

also holds when pushing forward or pulling

backward. For ease of reference, this inter-

dependence will be referred to as the ‘lifting DoF’.

Extending this concept, it may be seen that

adduction and abduction of the shoulder girdle can

be included. Abduction is connected to spine flexion,

particularly when the body bends down to pick

something up, as well as when pushing, and

shoulder adduction is connected with spine

extension during both lifting and pulling.

3.2 Semi-Active DoFs

It is possible to achieve a similar interdependency in

the exoskeleton by using a semi-active DoF. This is

a passive DoF driven by an active DoF.

PRACTICAL DESIGN OF FULL BODY EXOSKELETONS - Stretching the Limits of Weight and Power

135

Fig. 2 shows this concept schematically. Normally

the stator of a HAL-5 motor moves an arm or leg

segment while the axis is fixed to the exoskeleton

Figure 2: Placing bearings between the exoskeleton frame

(transparent) and the axis of a hip (or other) motor while

fixing the axis to a pulley, enables the pulley to drive a

second, passive joint that thus becomes semi-active.

trunk frame. The axis may instead be connected to a

pulley, and be allowed to rotate freely w.r.t. the

frame using a bearing. This pulley then drives an

otherwise passive joint through cables and a second

pulley at this passive joint, much like a common

cable actuation system. The torques in the active and

the semi-active DoFs are interrelated at any time,

thereby creating the desired interdependency.

The motor is, as in HAL-5, torque controlled

according to the BES of the wearer’s muscles. When

applying a semi-active DoF mechanism to assist

flexion of the exoskeleton spine, a torque controlled,

back-drivable hip motor produces the same force

balance in the exoskeleton as in the human lifting

DoF. Adding abduction of the shoulder as a second

semi-active DoF completes the robotic lifting DoF.

The moment in the human spine, however,

decreases when the wearer bends, because the

moment arm between the load and the spine, and the

moment arm between the load and the hips change

unequally. To achieve this effect with the robot, a

four-link mechanism between each hip motor and

the leg it drives may be used to increase the moment

at the robot’s legs when the legs are flexed, and thus

relatively decrease the moment in the spine.

Considering the ways we lift objects it may also

be seen that in a similar way elbow and wrist

actuation can be connected using a semi-active DoF,

thereby simplifying design.

3.3 Exo-Spine Structure

Using semi-actuation it is possible to support all

three DoFs of the spine from both hips. First, just as

the two hip moments in a human body act as one

moment on the trunk, the two axes of the two hip

motors can be connected in order to let the total

torque in this single axis act on the robot trunk.

Next, making sure that the exo-spine has a

straight neutral position, any deviation should cause

a moment that tends to restore the neutral position.

This is applicable to each spinal DoF because when

the wearer lifts something support is required in all

directions in order to pick it up while being rotated,

bend sideways or while using one hand. In effect, all

three DoFs are connected into a unified DoF that

tends to restore the neutral position.

Figure 3: Spine structure composed of vertebrae and links,

some overlaid by schematic equivalents, in fully bent and

straight positions. It extends when bending forward to

accommodate human spine flexion.

Figure 4: Side bending (a) and rotation (b) (top view) of

the spine structure. Beams were added for clarification.

A suitable spine-like mechanism is shown in

Figs. 3 and 4. The details of the design are beyond

the scope of this paper, but it can be seen that all

three DoFs of the human spine are provided. Several

vertebra-like segments and links are interconnected

by ball joints, while two synthetic cables (not

shown) connected to the axes of the hip motors pull

BIODEVICES 2010 - International Conference on Biomedical Electronics and Devices

136

the structure towards the neutral position. The cables

continue from the spine upwards to support against

shoulder girdle abduction during lifting and pulling.

The connected axes of the hips are balanced in

only one direction. In the other direction a torque-

clutch locks the axes of the motors to the frame,

depending on the direction of the combined torque

produced by the motors and the cables.

Given sufficient motor torque, the exo-spine is

analyzed using FEM to be strong enough to support

80kg at 24cm in front of the center of the wearer,

which would relieve most of a nurse’s load.

Moreover, all forces applied to the wearer pull

towards the neutral position, and hyperexten-sion is

blocked, making it safe to use. The exo-spine

requires no extra actuators, provides gravity

compensation, and supports pulling and pushing as

required, utilizing practical human mechanics.

Figure 5: Schematic CAD model of the trunk and one arm.

Robotic joints 1’- 4’ do not align with the sternocla-

vicular joint (1) and the glenohumeral joint (2-4). Motions

between the human and the robot at the fastening

equipments are accommodated by extra passive DoF.

4 OTHER SOLUTIONS

4.1 Intentional Misalignment

Two passive DoF added after each active arm joint

as proposed by Schiele and Van der Helm (2006) not

only facilitate unavoidable misalignments between

the robot and human joints, they also allows larger

misalignments. Using this concept, offsetting the

three robot joints at the shoulder w.r.t. the

glenohumeral joint would create space for the

wearer’s shoulder girdle to move upward, without

the need to actuate such a DoF. This increases the

RoM of the arm, since raising the arm beyond about

90 degrees involves upward motion of the girdle.

Fig. 5 illustrates this. Robot axes 1’, 2’, 3’, and 4’

have been misaligned intentionally w.r.t. their

human counterparts. Axis 1’ facilitates girdle

abduction, supported by the same cables as the exo-

spine. It has a large misalignment, but a small RoM,

allowing it to function as desired. Due to the added

passive DoFs the moment effectively put on the

human arm for a constant motor torque differs by a

few percent according to the posture. This is often

disrupting for machines, but is not sufficient to

influence the wearer of an exoskeleton.

4.2 Gravity Compensation

The human arm has a large workspace, but only

some of the robot arm’s DoFs can be active, and

they must always be positioned such that they

compensate gravity forces when required. In general,

people often lift objects with their elbows kept down

as much as possible, e.g. as farmers do (Nevala-

Puranen 1995). Assuming this, an arrangement of

passive and active DoFs as shown in Fig. 5 would let

the output of each motor compensate gravity as

much as possible. This is because the axes of the

motors are always perpendicular both to the gravity

forces as well as to the line connecting the center of

the motor to the point where the load is applied.

(a) (b)

Figure 6: Single stance phase during stair climbing with

HAL (a). A model (b) shows how the force (F

) at the

center of mass (CM) and the floor reaction force (FRF)

create a moment at the hip joint of the stance leg, M

hip

At the hips, forces acting from the back part of the

suit are supported by the suit’s legs. Standing on

both legs is no problem, but when walking, during

the single stance phase as shown in Fig. 6, a large

moment, about 2Nm/kg, is developed around the hip

of the stance leg. Because nurses on occasion need

PRACTICAL DESIGN OF FULL BODY EXOSKELETONS - Stretching the Limits of Weight and Power

137

to walk when lifting, these moments should be

supported. However, it is also desirable that the

wearer be able to abduct the leg. Passive joints that

only allow abduction of the legs would solve this.

5 DISCUSSION

Even with the proposed solutions the variety of

motions that can be performed with HAL is still less

than without, and limitations remain. Gravity

compensation is limited to generic postures, some

useful DoFs, such as inner rotation of the arm, are

not augmented, and the full RoMs are not achieved.

Even so, in most working situations there are

several postures available to the worker by which

the task’s goals can be achieved, and the wearer may

adapt his motion to utilize postures for which HAL

provides the most support. Since this is available in

postures humans use extensively it is very likely that,

although it should be confirmed by further research,

at any time at least one good posture can be attained.

Therefore, HAL would be valuable in a human

environment and the proposed practical design

approach thus achieved its goals. In addition, it is

expected that further practical, human characteristics

may be exploited to simplify design.

We believe that a similar practical focus may be

applied to other fields where humans and machines

meet cooperatively, such as rehabilitation, to yield

new improvements. A practical approach could

unveil solutions that enable patients to perform

particularly those motions needed for daily activities.

ACKNOWLEDGEMENTS

This work was supported in part by the Global COE

Program on “Cybernics: fusion of human, machine,

and information systems”.

REFERENCES

Aida, T., Nozaki, H. & Kobayashi, H., 2009.

“Development of Muscle Suit and Application to

Factory Laborers” Proc. IEEE Int. Conf. on

Mechatronics and Automation, China, pp.1-5.

BBC News, 2008. "US army develops robotic suits", April

16 2008, retrieved August 31, 2009, <news.bbc.co.uk

/2/hi/7351314.stm>.

Colombo, G., Joerg, M., Schreier, R. & Dietz, V., 2000.

“Treadmill training of paraplegic patients using a

robotic orthosis,” J. Rehabil. Res. Develop., vol. 37,

no. 6, pp. 693–700.

Dollar, A.M., & Herr, H., 2008. “Lower Extremity

Exoskeletons and Active Orthoses: Challenges and

State of the Art,” Transactions on Robotics, vol. 24(1).

Folgheraiter, M., Bongardt, B., Schmidt, S., De Gea, J.,

Albiez, J. & Kirchner, F., 2009. “Design of an Arm

Exoskeleton Using a Hybrid Model- and Motion-

Capture-Based Technique,” Interfacing the Human

and the Robot Workshop, ICRA.

Hidler, J.M. & Wall, A.E., 2005. “Alterations in muscle

activation patterns during robotic-assisted walking,”

Clin. Biomech., vol. 20, pp. 184–193.

Liszka, M.S., 2006. “Mechanical Design of a Robotic Arm

Exoskeleton for Shoulder Rehabilitation,” M.S. thesis,

Univ. of Maryland.

Mulas, M., Folgheraiter, M. & Gini, G., 2005. “An EMG-

controlled Exoskeleton for Hand Rehabilitation” Int.

Conf. on Rehabilitation Robotics, Chicago, IL, USA.

Nevala-Puranen, N., 1995. “Reduction of farmers' postural

load during occupationally oriented medical reha-

bilitation,” Applied Ergonomics, 26(6), pp. 411-415.

Perry, J.C. & Rosen, J., 2006. “Design of a 7 Degree-of-

Freedom Upper-Limb Powered Exoskeleton,” IEEE

Conf. on Biomedical Robotics and Biomechatronics.

Rosen, J., Perry, J.C., Manning, N., Burns, S. &

Hannaford, B., 2005. “The human arm kinematics and

dynamics during daily activities – toward a 7 DOF

upper limb powered exoskeleton,” Proc. 12th Intl.

Conf. on Advanced Robotics, ICAR '05, pp. 532-539.

Sasaki, D., Noritsugu, T., Takaiwa, M., & Yamamoto, H.,

2004. “Wearable power assist device for hand

grasping using pneumatic artificial rubber muscle,”

13th IEEE Int. Workshop on Robot and Human

Interactive Communication, ROMAN, pp. 655-660.

Schibye, B., Faber Hansen, A., Hye-Knudsen, C.T.,

Essendrop, M., Bocher, M. & Skotte, J., 2003,

“Biomechanical analysis of the effect of changing

patient-handling technique,” App Ergonom, 34, pp.

115-123.

Schiele, A., & Van der Helm, F.C., 2006. “Kinematic

design to improve ergonomics in human machine

interaction”, IEEE Trans Neural Syst Rehabil Eng,

14(4), 456-69.

Suzuki, K., Kawamura, Y., Hayashi, T., Sakurai, T.,

Hasegawa, Y. & Sankai, Y., 2005, "Intention-Based

Walking Support for Paraplegia Patient," Int. Conf. on

Systems, Man and Cybernetics, Hawaii, pp.2707-2713.

Toyama, S., 2009. “Wearable Agri Robot,” Toyama Lab,

retrieved August 31, 2009, <www.tuat.ac.jp/

~toyama/research_assistancesuitE.html>.

Tsagarakis, N. & Caldwell, D., 2003. “Development and

control of a soft-actuated exoskeleton for use in

physiotherapy and training,” Autonomous Robots, vol.

15, pp. 21–33.

Vieira, E.R. & Kumar, S., 2004 “Working postures: a

literature review,” J Occup Rehabil, 14(2), pp. 143-59.

Yamamoto, K., Hyodo, K., Ishi, M. & Matsuo, T., 2002.

“Development of power assisting suit for assisting

nurse labor,” JSME Int. J., Series C: Mechanical

Systems, Machine Elements and Manufacturing, v 45,

n 3, pp. 703-711.

BIODEVICES 2010 - International Conference on Biomedical Electronics and Devices

138