Handling Comparison between a Human and a Patient Simulator for

Nursing Care Related Physical Human-robot Interaction

Christian Kowalski

, Pascal Gliesche

, Conrad Fifelski-von Böhlen

, Anna Brinkmann

and Andreas Hein

1,2

OFFIS Institute for Information Technology, 26121 Oldenburg, Germany

Carl von Ossietzky University, 26129 Oldenburg, Germany

Keywords:

Human-robot Interaction, Nursing Care, Physical Relief.

Abstract:

The occurrence of musculoskeletal diseases among nursing staff leads to an early withdrawal from the profes-

sion, which reinforces the already existing lack of caregivers. To counteract this problem, we would like to

provide physical relief through robotic assistance at the bedside. However, the problem arises that for safety

reasons robotic assistance concepts should not be tested on humans at ﬁrst. In this case, patient simulators

with an average person’s weight can function as a substitute. For the best results, both cases should behave

very similarly when evaluating robot assistance concepts so that the transfer from patient simulator to human

is small and therefore no major adjustments need to be made. To measure this potential difference, we have

compared the handling properties of both cases in this paper. We examined force measuring platform data

while a nurse mobilized an 80 kg human and a patient simulator from the back to the side. The experimental

results show that moving a patient simulator is more physically demanding compared to moving a human

with similar weight and that conventional collaborative lightweight robots are able to push and move a patient

simulator that is weighing far higher than the robot’s actual payload suggests.

1 INTRODUCTION

Nursing shortage is an already existing problem in

many countries. Demographic change means that

more and more people of advanced age have to be

cared for. In Germany in particular, the number

of nursing staff is also tending to decline (Schulz-

Nieswandt, 2018). The rising number of people in

need of care is being offset by a decreasing number

of qualiﬁed nursing staff, so that according to fore-

casts there will be a shortage of more than 450.000

caregivers by 2050 (Ehrentraut et al., 2015). One of

the reasons for the decrease in workforce size is the

early withdrawal from work due to musculoskeletal

disorders (Trinkoff et al., 2003). Especially nursing

activities at the bed have an enormous inﬂuence on

the load acting on the caregivers’ spines (Jäger et al.,

2013).

Common physical support tools are only of lim-

ited help in this regard. For instance, the time con-

suming usage of patient lifters is usually limited to the

transfer to or into the bed and thus does not actively

promote the patient’s mobility. A further possibil-

ity of relief is the augmentation of the nursing staff’s

Figure 1: Setup for the support of nursing care bed activities

using a bedside mounted KUKA LBR iiwa 7 R800 (a), an

AMTI AccuPower portable force measuring platform (b),

a rescue dummy functioning as a patient simulator with a

weight of 80 kg (c) and a multi depth camera system (d).

strength by exoskeletons worn on the body (Liu et al.,

2016; Taal and Sankai, 2011). However, an ex-

oskeleton has the disadvantages that it must be put

on before use and it is not able to replace the help of

a second nurse, who is often called in for physically

Kowalski, C., Gliesche, P., Böhlen, C., Brinkmann, A. and Hein, A.

Handling Comparison between a Human and a Patient Simulator for Nursing Care Related Physical Human-robot Interaction.

DOI: 10.5220/0010322006050612

In Proceedings of the 14th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2021) - Volume 5: HEALTHINF, pages 605-612

ISBN: 978-989-758-490-9

605

demanding activities. As mentioned before, calling

a second person for help is often no longer possible

due to lack of personnel. Therefore, we envision to

develop a robotic assistance system that acts as a sub-

stitute for the otherwise missing second caregiver. For

this goal, we designed a setup in previous work where

a KUKA LBR iiwa 7 R800 was mounted to an actu-

ated care bed which is surrounded by a multi depth

camera arrangement (Fifelski et al., 2018) consisting

of Microsoft Azure Kinect 3D cameras and a force

measuring platform (Kowalski et al., 2020). Addi-

tionally, as the system is not certiﬁed according to

the Medical Device Directive (MDD of the European

Community), a rescue dummy with the weight of an

average person functions as a patient simulator in this

context (see Fig. 1). In our opinion, three key aspects

need to be researched in order to develop a support

concept for the relief of bedside nursing staff in car-

rying out nursing activities:

1. Force transmission concepts to a potentially vul-

nerable patient, serving a multitude of manipula-

tions and body anatomies.

2. Detection of suitable spots to interact with the pa-

tient.

3. Operation of the robot in orchestration with the

caregiver for an optimal interaction during task

execution.

In this work, we mainly want to discuss boundary

conditions that can lead us to the solution of the three

aforementioned key aspects. An extremely relevant

point is the testing of robotic support concepts in

real life. However, from an ethical and safety point

of view, it is more advantageous to initially test the

robot’s performance on patient simulators with hu-

man weight instead of real humans. The question

whether and to what extent the handling of such a

simulator differs from a real human being has to be

answered, which will be examined in more detail in

this work. Subsequently, we want to take a closer look

at the potential of a conventional light weight robot

manipulator such as the KUKA LBR iiwa 7 R800 to

move a patient simulator on its own. These robots are

not necessarily built to withstand the weight of people

during nursing activities.

After the presentation of the related work, we will ﬁrst

present the results of the aforementioned nursing ac-

tivity analysis. Then, the two different experiments

including the comparison between a human and a pa-

tient simulator and the force transmission to move a

patient are conducted and discussed.

2 RELATED WORK

The intentional contact between robot and human

with additionally controlled force transmission is a

rather unusual topic, which is why the number of

publications for this particular subject is quite sparse,

even in the ﬁeld of physical Human-Robot Interaction

(pHRI). In the health care domain, Erickson et al. pre-

sented a work on washing patients using a PR2 robot

with the help of capacitive sensing instead of vision

or force feedback (Erickson et al., 2019a). The robot

was able to follow and clean a human arm by follow-

ing its contour while maintaining a force under a cer-

tain threshold. The capacitive sensing neural network

model was trained to estimate the relative position of

the closest point on a person’s limb surface. For mo-

tion control, a high level Cartesian controller was used

to provide joint values to the low-level proportional-

integral-derivative (PID) controllers of the robot ac-

tuators. Only a few researchers have worked towards

the goal of washing a patient and in most cases the

they did not tackle the washing problem directly but

rather developed exoskeletons (Satoh et al., 2009) or

bath water control systems. King et al. again devel-

oped a robot that wipes off debris of a human’s up-

per arm, forearm, thigh or shank lying in a bed using

a compliant force-controlled wiping motion without

tracking but with the help of an operator (King et al.,

2010).

Another group of works deals with the aspect

of developing robots for the purpose of massaging.

Here, the contact between robot and human with si-

multaneous application of a predeﬁned force is in-

tended, whereby the contact forces with the soft tis-

sue of the human are particularly difﬁcult to assess.

Except for the bones in the human body, everything

else is deﬁned as soft tissue and can be distinguished

by their different characteristics. The Young’s Modu-

lus of typical soft tissue is relatively low with a value

of around 1 MPa (Akhtar et al., 2011) and its model

can be described as a multilayered, anisotropic, vis-

coelastic, inertial, plastic and non-stationary environ-

ment (Golovin et al., 2014). In contrary, the Young’s

Modulus of skin can range from 5 kPa to 140 MPa

(Akhtar et al., 2011). This shows, that it is difﬁcult

to grasp the properties in advance to a contact and

that they also most likely vary from person to person

based on the body composition, muscle contraction

and many other factors, once again showing the com-

plexity of soft tissue contact scenarios. Nevertheless,

Golovin et al. incorporated a control method includ-

ing position and force to perform the task of massag-

ing (Golovin et al., 2014). In most pHRI use cases,

a compliant robotic behaviour is desirable, which is

why impedance control is often the ﬁrst choice in this

HEALTHINF 2021 - 14th International Conference on Health Informatics

606

area (Hogan, 1985; De Santis et al., 2008; Haddadin

et al., 2008).

In the context of medicine and surgery, the appli-

cation of force directly on humans by robotic assis-

tance systems is not a novelty, but the forces applied

are relatively small compared to the forces occurring

during the execution of nursing tasks (Peirs et al.,

2004; Ho et al., 1995). Another ﬁtting area of work

in robotics is the manipulation of objects in the envi-

ronment by pushing, which is usually the method of

choice when the target object is too big or too heavy to

grasp. Just like in the aforementioned literature, ma-

nipulation by pushing is not trivial due to the many

geometrical and physical properties associated with

the robot’s surroundings. In general, for planning and

control either a forward model or an inverse model is

used to predict the next state based on an action of

the current state or to compute the action that changes

the current state to a desirable target state (Stüber

et al., 2020). There are many different approaches to

this topic, ranging from deep (reinforcement) learning

(Peng et al., 2018; Byravan and Fox, 2017; Ehrhardt

et al., 2017), data-driven (Stüber et al., 2018; Ridge

et al., 2015), analytical (Lee et al., 2015; Dogar and

Srinivasa, 2011) to physics engine (Zhu et al., 2017)

based methods. Although the pushing methods pre-

sented so far cover a broad ﬁeld, to our knowledge

they have not yet been applied in the context of nurs-

ing, which adds a whole new layer of complexity due

to safety reasons.

3 APPROACH

3.1 Nursing Activity Investigation and

Physical Load Limits

In the beginning, a small focus group meeting was

held with four people attending who had a nursing

background. The reason for the meeting was, on the

one hand, to identify the everyday nursing activities

at the bed, which require physical effort and on the

other hand, to explore cooperative activities, since

in some cases the activity cannot be easily managed

alone. The activities were also carried out in an ex-

emplary fashion and recorded using the Azure Kinect

3D depth cameras for later analysis. The activities

determined were then compared with the literature to

obtain a complete representation. Then, the activi-

ties were compared with the ones used in a study by

Jäger et al. to determine the loads on the lumbar spine

with the help of a biomechanical model (Jäger et al.,

2013). In Table 1 these values are compared with the

maximum recommended lumbar load for healthy and

back-friendly working (Jäger, 2019). It is noticeable

that the execution of most nursing care bed activities

exceeds the load limits and therefore has a negative

impact on the musculoskeletal system. Another

aspect, which is of great relevance in this context, is

the consideration of the maximum forces that can be

applied to the human body. Due to the fact that the

intended transmission of force using robots is rarely

carried out, no values have yet been determined for

this application. However, it is possible to fall back

on safety values for collisions with robots for the

time being. Table 2 shows the maximum permissible

forces in Newtons per body region, which are derived

from DIN ISO/TS 15066 (ISO, 2017).

To validate the physical relief, in our case a force

measuring platform is placed in front of the bed in the

nurse’s work area. This does not allow a direct com-

parison with the results of the biomechanical model

of Jäger et al. (Jäger et al., 2013), but it is possible to

have a look at the measured ground reaction forces of

the nurse to draw conclusions from these data. Fur-

thermore, a good picture of the overall force distribu-

tion can be generated with the torque data from each

individual robot joint so that physical relief becomes

quantiﬁable.

4 EXPERIMENTS AND RESULTS

4.1 Handling Comparison between a

Human and a Patient Simulator

The creation of a realistic test scenario is an impor-

tant factor for many areas of robotics. While many

scientiﬁc papers deal with the generalization of robot

behavior and try to represent the real world in simula-

tions, real data remain irreplaceable for testing pur-

poses for the time being. Especially manipulation

tasks are very difﬁcult to reproduce in simulations

due to the complexity caused by the direct contact

with all associated physical parameters (Peng et al.,

2018). This makes testing and data collection in the

real world all the more important. In the case of pHRI,

however, this turns out to be problematic, since dur-

ing development of a robot’s behaviour the collec-

tion of data directly on humans should be circum-

vented for reasons of safety and ethics. This problem

has been recently recognized for assistance robotics

and there exist approaches to collect data directly on

human models from robots in simulation (Erickson

et al., 2019b). In addition to simulation, data collec-

tion in the nursing context in the real world would

Handling Comparison between a Human and a Patient Simulator for Nursing Care Related Physical Human-robot Interaction

607

Table 1: Mean values and ranges of compressive force on the lumbosacral disc for three different execution modes of nine

nursing activities based on the results of Jäger et al. (Jäger et al., 2013). The appropriate force limit starts at 4.1 kN for 20

year old women and decreases down to 1.8 kN for 60+ year old women. For men the limits range from 5.4 kN to 2.2 kN

(Jäger, 2019).

Nursing activity Conventional Optimized Small aids)

a. Raising from a lying to a sitting position 3.4 (1.8 - 5.4) 2.3 (1.9 - 2.9) n.a

b. Elevating to a sitting position at the bed’s edge 5.0 (3.3 - 6.2) 2.7 (2.0 - 3.6) n.a.

c. Moving to the bed headboard with nurse at bed’s side 6.7 (5.6 - 8.0) 5.4 (3.7 - 6.5) 2.8 (2.3 - 3.2)

d. Moving to the bed headboard with nurse at bed’s head 5.7 ( 2.8 - 8.9) 2.5 (2.0 - 3.0) 2.4 (2.2 - 2.8)

e. Moving sidewards 4.9 (3.3 - 5.8) 2.6 (2.0 - 3.4) 1.9 (1.6 - 2.2)

f. Raising the bedhead 4.3 (3.8 - 5.4) 4.1 (3.5 - 5.2) n.a.

g. Assisting with a bed-pan 4.2 (2.6 - 6.5) 2.6 (1.6 - 3.3) n.a.

h. Moving from the bed into a chair 5.1 (3.8 - 6.5) 3.7 (2.3 - 4.4) 3.1 (1.6 - 5.3)

i. Raising from sitting to an upright position 4.9 (3.8 - 6.4) 2.5 (1.9 - 3.1) n.a.

Table 2: Body contact force limits based on (ISO, 2017).

Maximum permissible

Body region contact force [N]

Head 130

Face 65

Neck 150

Back 210

Shoulders 210

Chest 140

Abdomen 110

Pelvis 180

Upper arms 150

Forearms 160

Hands 140

Thighs 220

Calves 210

Figure 2: Process of the experiment to determine the dif-

ferences between handling either a patient simulator or a

human. A nurse is standing on a force measuring platform

(a) while turning the patient to each side (b-c) during the

process of moving towards the bed’s headboard.

also be conceivable with patient simulators. Since no

other work has yet made a comparison with regard to

the forces acting between patient simulators and hu-

mans, we will conduct the research for this particu-

lar topic with the infrastructure described before. For

the purpose of this comparison, the strenuous activ-

ity of moving the patient to the bed’s headboard when

standing at the bed’s side was performed by a care-

giver while standing on a force measuring platform

(see activity c in Table 1 and Fig 2). The activity was

performed ﬁve times with a 80.5 kg person and a 80

kg patient simulator. The process is divided into two

steps: The lying person or the lying patient simulator

is ﬁrst turned to the side in the direction of the nurse

and is moved slightly towards the headboard when the

person is put back on his back. Then the same move-

ment is repeated but this time away from the nurse.

The entire process is also recorded by a depth cam-

era, so that the times of the two turning processes can

be tracked exactly for a precise analysis. The result-

ing forces of one pass can be seen in Fig. 3. In all

ﬁve passes, the two body turning activities were anal-

ysed both individually and together for every axis,

which can be seen in Table 3. In particular, the arith-

metic mean value, standard deviation, minimum and

maximum values were calculated for further inspec-

tion. If one compares the two turn activities with each

other, it becomes clear that the second turning event

of the patient requires less effort in the case of the

human and more effort in the case of the patient sim-

ulator. This becomes particularly obvious by looking

at the forces in the direction of the z-axis which is

5.27 times higher. Also, the overall standard devia-

tion is slightly larger during the ﬁrst turn activity. Fur-

thermore, the minimum values when moving a human

are approximately the same during both turns with a

small difference of 8.8, the values while moving the

patient simulator are much further apart where the av-

erage value for the ﬁrst turn is -130 while the second

turn has a value of only -24.1. In both scenarios, the

ﬁrst turn activity has a higher value for the maximum

values, but the difference is greater for the human with

a value of 65 at the ﬁrst turn while the difference for

the patient simulator is only 18.37. As expected, the

maximum values also show the highest peak load of

310.8 on average for the human and 388.7 for the pa-

tient simulator in the direction of the x-axis of the

force measuring platform, which is most likely due to

HEALTHINF 2021 - 14th International Conference on Health Informatics

608

Figure 3: Raw force data (x, y and z axis) recorded while a nurse moves a human or a patient simulator towards the bed’s

headboard. The activity mainly consists of two turn movements (areas marked in red) which have been annotated using the

data of a depth camera recording.

Table 3: Results of the comparison between the handling of

a patient simulator and a human during the nursing activ-

ity of moving a patient towards the bed’s headboard in two

individual steps.

Human Patient Simulator

Turn: 1st 2nd 1st 2nd

∅ F [N] 51.7 39.2 66.6 107.6

∅ F_x [N] 112.9 52 156.4 135.3

∅ F_y [N] 21.5 30.7 19.4 60.2

∅ F_z [N] 20.6 35 24.1 127.2

∅ SD [N] 62.2 55.7 87.6 57

∅ SD_x [N] 85.8 49.2 124 43.6

∅ SD_y [N] 66 80.7 68.8 61.2

∅ SD_z [N] 34.8 37.2 70 66.1

∅ Min [N] -93.8 -85.0 -130 -24.1

∅ Min_x [N] -21.8 -24 -59.4 35.1

∅ Min_y [N] -174.4 170.2 -162.8 -83.6

∅ Min_z [N] -85.3 -60.9 -167.8 -24.7

∅ Max [N] 221.6 156.6 265 246.63

∅ Max_x [N] 310.8 155.9 388.7 214

∅ Max_y [N] 188.4 178.9 220.7 218.8

∅ Max_z [N] 165.7 134.9 185.4 307.2

the leverage when pulling the patient during the ﬁrst

turn. This high value can also be found in the over-

all force, especially in the x-axis component. This is

also where the two most important statements regard-

ing the validity of the data can be found: ﬁrst, the

x-axis component of the force data has very high val-

ues during the execution of the task. Second, turning

towards the nurse - i.e. the ﬁrst part of the activity -

is more strenuous in both scenarios. Third, moving

the patient simulator is more strenuous than moving

the human being of an almost identical weight. On

average, turning towards the nurse is 1.28 times more

difﬁcult using the patient simulator according to the

measurement and even 2.74 times more difﬁcult when

during the turn away activity. Interesting at this point

is the difference in the load peaks, which are given

by the maximum values. During the ﬁrst turn activ-

ity we measured a 1.2 times higher maximum force

when handling the patient simulator, during the sec-

ond turn activity it is even 1.57 times higher. It can

be concluded from the results that in the process of

placing the patient on his side, the patient simulator

with an almost identical weight cannot reproduce the

kinematics, material characteristics or loads of a real

human being. This experiment suggests that for a

test environment similar to that of a real human be-

ing, the patient simulators either need to be equipped

with better mobility or they need a lower weight in or-

der to map the potential load forces of a person with

more weight. It must be said, however, that the ex-

periment is limited to the performance of one speciﬁc

nursing activity and the results may vary signiﬁcantly

for other activities.

4.2 Maximum Robot Load for Physical

Human-robot Interaction

Another important aspect, which is necessary for the

investigation of pHRI in the ﬁeld of physical assis-

tance in care, is the payload or the potential of the

robots to move larger masses. It is well known that

robots are capable of moving large masses. However,

the potential maximum payload depends on the de-

sign, the conﬁguration in respect to the patient and

the maximum torques at the relevant joints of the

robot. In nursing, however, it is also necessary to

have enough space for the collaborating caregiver.

The keyword "collaborating" is particularly impor-

tant here, since most collaboration robots are built

smaller and have lower payloads than common indus-

trial robots. In our example setup, an iiwa manipula-

tor with a maximum payload of 7 kg is used. How-

ever, for health care support, the robot has to cope

with the patients’ weight to provide physical relief.

Handling Comparison between a Human and a Patient Simulator for Nursing Care Related Physical Human-robot Interaction

609

Figure 4: The setup of the robotic load experiment where

the robot tries to push the 80 kg weighing patient simulator

without any external help. The pushing starts at about 2.4

seconds and ends after 12.2 seconds. The torque for every

joint during the execution is measured.

For this case we have carried out a joint load test ex-

periment, where the robot manipulator should inde-

pendently move the patient simulator by pushing it

within the bed. To be more precise, the robot’s start

position q

start

(t) and goal position q

goal

(t) in joint

space have already been deﬁned in advance so that

we only have to deal with the Cartesian movement

between these positions. In addition, we are only

considering the translational component of the move-

ment, breaking it down to a one dimensional motion

along the Y-axis relative to the robot’s base frame. For

the experiment, the end effector presses on the upper

arm of the patient simulator and thus moves it side-

ways by an amount of about 10 cm without additional

help (see Fig. 4). The resulting external torques at the

individual joints were observed over time (see Fig. 5).

The experiment was repeated 8 times and in all ex-

periments it was possible to move the 80 kg patient

simulator by about 10 cm without external help. The

experiment carried out reveals two important points:

ﬁrst, the robot’s payload is not decisive for the max-

imum applicable force to move masses and second,

it is necessary to optimize both the conﬁguration to

support without disturbing the caregiver and the joint

loads of the robot for maximum exploitation of the

push potential to maintain τ

min

(t) ≤ τ

τ ≤ τ

max

(t) due

to the robot’s maximum allowed joint torques in New-

ton meters, being 176, 176, 110, 110, 110, 40 and

40 for the used robot manipulator beginning from the

robot’s ﬁrst joint (base) to the last joint (end effec-

tor). This is very important because in the context

of nursing care, any robotic support movement will

have to deal with the problem of applying a prede-

ﬁned force on one or more body parts of the patient

to cooperate with the nurse during the task execution

to ﬁnally provide physical relief. The complexity of

the trajectories while applying the forces can arbitrar-

Figure 5: Visualization of the joint torques during the

robotic load experiment where the robot applies a force to

move the 80 kg weighing patient simulator. Mean torques

of every joint and lower and upper error are visualized.

ily increase or decrease and is not dependent on the

actual force transmission itself, except in relation to

the force limit values, which must be adjusted regard-

ing the selected body part as shown in Table 2. As

already stated before, our system uses the FRI in or-

der to achieve a control loop frequency of up to 1

kHz. However, this also limits the obtainable robot

information so that only the individual external joint

torques can be acquired. For nursing care, it would

be best to make assumptions about the Cartesian end

effector forces without using any additional sensors.

For this particular case, it is possible to predict the

forces by using the relationship between applied end

effector wrenches and applied forces and torques to

the joints as in (Siciliano and Khatib, 2016)

τ = J

(

)

f , (1)

where τ

τ is the forces and torques vector for a robotic

manipulator of n degrees of freedom (DOF), J

is the

transposed Jacobian matrix and f

f is the end effector

force vector. To get the actual Cartesian end effector

forces it is possible to make use of the Moore-Penrose

inverse to ﬁnally get

f = (J

(

)

−1

τ. (2)

Multiplying this result with the Jacobian gives us the

Cartesian end effector forces in its local frame. In

real care scenarios using a robot, the support move-

ment should make use of the force measurements to

constantly update the position along a predeﬁned tra-

jectory to maintain the applied force below a desired

threshold depending on the individual body part, the

values for each can be found in Tab. 2 (ISO, 2017).

HEALTHINF 2021 - 14th International Conference on Health Informatics

610

5 CONCLUSIONS

In this paper we were able to collect important aspects

for the approach to the topic of pHRI in the care do-

main. There is a general need for physical relief in

care. For this relief through robotic assistance, how-

ever, a force application directly or indirectly on hu-

mans is necessary. Safety standards with values for

force limits depending on the body part do already ex-

ist but these were not created with the intention of pro-

viding relief in care and are currently only means to an

end. It requires a systematic creation of care-related

force limits. When testing care-relevant robotic sup-

port movements, initial experiments using humans is

not desirable and one should switch to patient sim-

ulators for this particular task. In the present paper,

however, it could be shown that there is a mismatch

between patient simulators and humans, which must

either be taken into account or developments in this

ﬁeld must take place so that simulators become more

similar to humans with a suitable weight, material and

mobility. Finally, we were able to show that even col-

laborative lightweight robots can apply enough force

to independently move an 80 kg patient simulator in

bed and are thus also suitable for nursing activities.

6 FUTURE WORK

The presented work should serve as a basis for the

ﬁeld of pHRI for nursing care and should also show

that despite existing gaps in the framework condi-

tions, there is a potential for force relief of caregivers

by collaborative robots. In future work, we will fo-

cus on directly supporting caregivers using robots and

on measuring and comparing the degree of potential

physical relief. For this, the three main difﬁculties in

this complex project mentioned at the beginning have

to be considered more intensively in follow-up work.

On the one hand, an additional assessment of physi-

cal properties may possibly provide an advantage in

the transmission of force. On the other hand, nursing

activities at the bed are such highly complex activi-

ties that this problem should perhaps be handled by a

robot controller learned through reinforcement learn-

ing rather than using a handcrafted controller. Over-

all, there are still many areas where the present system

can be further improved and used for research. We

envision a system which ensures signiﬁcant physical

relief through human-robot interaction and coopera-

tion while maintaining safety standards with regard

to maximum force limits dependent on the patient’s

condition.

ACKNOWLEDGEMENTS

This work was funded by the German Ministry

for Education and Research (BMBF) within the re-

search project Nursing Care Innovation Center (grant

16SV7819K).

REFERENCES

Akhtar, R., Sherratt, M. J., Cruickshank, J. K., and Derby,

B. (2011). Characterizing the elastic properties of tis-

sues. Materials Today, 14(3):96–105.

Byravan, A. and Fox, D. (2017). Se3-nets: Learning rigid

body motion using deep neural networks. In 2017

IEEE International Conference on Robotics and Au-

tomation (ICRA), pages 173–180. IEEE.

De Santis, A., Siciliano, B., De Luca, A., and Bicchi, A.

(2008). An atlas of physical human–robot interaction.

Mechanism and Machine Theory, 43(3):253–270.

Dogar, M. and Srinivasa, S. (2011). A framework for push-

grasping in clutter. Robotics: Science and systems VII,

Ehrentraut, O., Hackmann, T., Krämer, L., and Schmutz,

S. (2015). Zukunft der Pﬂegepolitik: Perspek-

tiven, Handlungsoptionen und Politikempfehlungen.

Friedrich-Ebert-Stiftung, Abteilung Wirtschafts-und

Sozialpolitik.

Ehrhardt, S., Monszpart, A., Mitra, N. J., and Vedaldi, A.

(2017). Learning a physical long-term predictor. arXiv

preprint arXiv:1703.00247.

Erickson, Z., Clever, H. M., Gangaram, V., Turk, G.,

Liu, C. K., and Kemp, C. C. (2019a). Multidimen-

sional capacitive sensing for robot-assisted dressing

and bathing. In 2019 IEEE 16th International Confer-

ence on Rehabilitation Robotics (ICORR), pages 224–

231. IEEE.

Erickson, Z., Gangaram, V., Kapusta, A., Liu, C. K., and

Kemp, C. C. (2019b). Assistive gym: A physics simu-

lation framework for assistive robotics. arXiv preprint

arXiv:1910.04700.

Fifelski, C., Brinkmann, A., Ortmann, S. M., Isken, M., and

Hein, A. (2018). Multi depth camera system for 3d

data recording for training and education of nurses. In

2018 International Conference on Computational Sci-

ence and Computational Intelligence (CSCI), pages

679–684. IEEE.

Golovin, V., Arkhipov, M., and Zhuravlev, V. (2014). Force

training for position/force control of massage robots.

In New trends in medical and service robots, pages

95–107. Springer.

Haddadin, S., Albu-Schaffer, A., De Luca, A., and

Hirzinger, G. (2008). Collision detection and reac-

tion: A contribution to safe physical human-robot in-

teraction. In 2008 IEEE/RSJ International Conference

on Intelligent Robots and Systems, pages 3356–3363.

IEEE.

Handling Comparison between a Human and a Patient Simulator for Nursing Care Related Physical Human-robot Interaction

611

Ho, S., Hibberd, R., and Davies, B. (1995). Robot assisted

knee surgery. IEEE Engineering in Medicine and Bi-

ology Magazine, 14(3):292–300.

Hogan, N. (1985). Impedance control: An approach to ma-

nipulation: Part i—theory.

ISO, D. (2017). Ts 15066: 2017-04. Robots and robotic

devices-Collaborative robots (ISO/TS 15066: 2016).

Jäger, M., Jordan, C., Theilmeier, A., Wortmann, N., Kuhn,

S., Nienhaus, A., and Luttmann, A. (2013). Lumbar-

load analysis of manual patient-handling activities for

biomechanical overload prevention among healthcare

workers. Annals of occupational hygiene, 57(4):528–

544.

Jäger, M. (2019). Die revidierten dortmunder richtwerte.

Zentralblatt für Arbeitsmedizin, Arbeitsschutz und Er-

gonomie, 69(5):271–289.

King, C.-H., Chen, T. L., Jain, A., and Kemp, C. C.

(2010). Towards an assistive robot that autonomously

performs bed baths for patient hygiene. In 2010

IEEE/RSJ International Conference on Intelligent

Robots and Systems, pages 319–324. IEEE.

Kowalski, C., Arizpe-Gomez, P., Fifelski, C., Brinkmann,

A., and Hein, A. (2020). Design of a supportive

transfer robot system for caregivers to reduce physi-

cal strain during nursing activities. In Studies in health

technology and informatics, volume 270, pages 1245–

1246.

Lee, G., Lozano-Pérez, T., and Kaelbling, L. P. (2015). Hi-

erarchical planning for multi-contact non-prehensile

manipulation. In 2015 IEEE/RSJ International Con-

ference on Intelligent Robots and Systems (IROS),

pages 264–271. IEEE.

Liu, C., Zhu, C., Liang, H., Yoshioka, M., Murata, Y.,

and Yu, Y. (2016). Development of a light wear-

able exoskeleton for upper extremity augmentation. In

2016 23rd International Conference on Mechatronics

and Machine Vision in Practice (M2VIP), pages 1–6.

IEEE.

Peirs, J., Clijnen, J., Reynaerts, D., Van Brussel, H., Heri-

jgers, P., Corteville, B., and Boone, S. (2004). A micro

optical force sensor for force feedback during mini-

mally invasive robotic surgery. Sensors and Actuators

A: Physical, 115(2-3):447–455.

Peng, X. B., Andrychowicz, M., Zaremba, W., and Abbeel,

P. (2018). Sim-to-real transfer of robotic control with

dynamics randomization. In 2018 IEEE international

conference on robotics and automation (ICRA), pages

1–8. IEEE.

Ridge, B., Leonardis, A., Ude, A., Deniša, M., and Sko

caj,

D. (2015). Self-supervised online learning of basic

object push affordances. International Journal of Ad-

vanced Robotic Systems, 12(3):24.

Satoh, H., Kawabata, T., and Sankai, Y. (2009). Bathing

care assistance with robot suit hal. In 2009 IEEE In-

ternational Conference on Robotics and Biomimetics

(ROBIO), pages 498–503. IEEE.

Schulz-Nieswandt, F. (2018). Stationaere Altenpﬂege und

«Pﬂegenotstand» in der Bundesrepublik Deutschland.

Peter Lang International Academic Publishers.

Siciliano, B. and Khatib, O. (2016). Springer handbook of

robotics. Springer.

Stüber, J., Kopicki, M., and Zito, C. (2018). Feature-based

transfer learning for robotic push manipulation. In

2018 IEEE International Conference on Robotics and

Automation (ICRA), pages 1–5. IEEE.

Stüber, J., Zito, C., and Stolkin, R. (2020). Let’s push things

forward: A survey on robot pushing. Frontiers in

Robotics and AI, 7:8.

Taal, S. R. and Sankai, Y. (2011). Exoskeletal spine and

shoulders for full body exoskeletons in health care.

Advances in Applied Science Research, 2(6):270–286.

Trinkoff, A. M., Lipscomb, J. A., Geiger-Brown, J., Storr,

C. L., and Brady, B. A. (2003). Perceived physi-

cal demands and reported musculoskeletal problems

in registered nurses. American journal of preventive

medicine, 24(3):270–275.

Zhu, S., Kimmel, A., and Boularias, A. (2017).

Information-theoretic model identiﬁcation and pol-

icy search using physics engines with applica-

tion to robotic manipulation. arXiv preprint

arXiv:1703.07822.

HEALTHINF 2021 - 14th International Conference on Health Informatics

612