SKILL ACQUISITION PROCESS OF A ROBOT-BASED AND A

TRADITIONAL SPINE SURGERY

Meike Jipp

Department of Psychology II, University of Mannheim, Schloss EO, Mannheim, Germany

Peter Pott

Laboratory for Biomechanics and Experimental Orthopeadics, Faculty of Clinical Medicine Mannheim,

Achim Wagner

, Essameddin Badreddin

Werner W. Wittmann

Keywords: Skill acquisition, System design, Man-machine interfaces

Abstract: Technological progress greatly revolutionizes medicine. Robots give the opportunity to reach greater

accuracy and thus improve the medical outcome. The results of a surgical intervention profoundly depend

on the robot system and on the training state of the operator. Since the learning of a surgical intervention

can be influenced by the complexity of the system, these interconnections are investigated with

psychological methods. Therefore, the skill acquisition process of a robot-based surgery is compared to a

traditional spine surgery. The usage of an appropriate robot shortens the learning curve of a spine surgery

due to a decreased complexity and reduces the impact of the surgeons’ psychomotor abilities on the

surgery’s outcome. For the design of a surgical system, different realizations must be tested in advance

regarding their cognitive workload to avoid training costs learning the operation of the system, afterwards.

The methods used are not restricted to surgical robotics.

1 INTRODUCTION

Medicine is one of the oldest sciences, as human

beings have always had a thirst for knowledge about

diseases and disorders. The development of

technology revolutionized medicine: in 1625 the

first microscope gave the opportunity to learn about

both bacteriology and chemistry

(http://techandmed.tripod.com/files/basicpages/histo

ry.htm), but only in the 21st century robotic systems

were approved for surgery

(http://electronics.howstuffworks.com/robotic-

surgery1.htm). Their capability to deal with large

amounts of data compensates for possible human

errors. Robots work more accurately and reduce the

convalescence time and possible traumata.

Nevertheless, with the use of robots, the surgical

process changes drastically and surgeons must

acquire additional skills. The design of the machine,

the man-machine interface, as well as the process of

surgery are limiting factors for the learning

complexity and are thus, responsible for clinical

results. However, the skill acquisition process with a

medical robot in particular is not well understood so

far. Considering the learning process in an early

stage of the system development may optimize the

outcome of the system and reduce the costs

drastically, which accumulate for the training of the

stuff. This interdisciplinary study between

engineering, medical, and psychological sciences is

one step to provide methods for the evaluation of the

interconnection between the learning process and the

Jipp M., Pott P., Wagner A., Badreddin E. and Wittmann W. (2004).

SKILL ACQUISITION PROCESS OF A ROBOT-BASED AND A TRADITIONAL SPINE SURGERY.

In Proceedings of the First International Conference on Informatics in Control, Automation and Robotics, pages 56-63

DOI: 10.5220/0001136900560063

 SciTePress

system design. Hopefully, it will help the designers

developing their technical systems to fit human

skills.

As an example of a technical system, the hand-

held device Intelligent Tool Drive (ITD) for bone

treatment (Pott, 2003) is investigated. In its first

application, it will be used for stabilizing two or

more vertebrae (arthrodesis) requiring the surgeon to

drill holes into the pedicles of the vertebrae in

question. Fig. 1 roughly explains a vertebra’s

anatomy.

Figure 1: Anatomy of a vertebra.

(a) refers to the pedicles, (b) to the vertebral body.

To drill holes manually into the pedicles is a

difficult task as there is no room for mistakes: first,

the drill is only about 1 to 2 mm smaller than the

pedicles. Additionally, the surgeon cannot constantly

check his or her work with X-rays or similar

procedures due to the amount of radiation on both

the surgical team and the patient. Thus, the surgeon

needs experience to combine the feedback of the

drill’s resistance with the knowledge of anatomy to

figure out whether he or she is drilling at the right

place. Although this is a routine surgery about 30%

of the screws are not set ideally, about 5-6% of these

require repeated surgery (Siebert, 2000).

The concept of the ITD allows the improvement

of this type of surgery, especially regarding the

drilling process’s accuracy. The desired drilling

position is planned before the intervention basing on

3D computer tomography. During the intervention,

the robot system measures the actual position of tool

and patient and controls the accurate positioning of

the drilling tool. In this way disturbances and

displacements by the surgeon, which holds the robot

in his or her hands, can be compensated.

Implementing the ITD will require the surgeons

to perform the surgery a completely different way.

With the ITD the surgeon must plan the drilling

trajectories and he or she must find anatomic

positions for the matching of the robot coordinates

with the patient coordinates. Little research has been

done on the process to acquire the skill to perform

surgery, although it is completely different applying

new tools, so that it is obvious that the new method

must be learnt. Is the skill acquisition process shorter

or qualitatively different when using a device

compared to a traditional surgery? What if the new

tool requires all of the surgeon’s attention and no

resources are left for the patient? Which are the

underlying abilities determining the learning

process? Can the impact of these abilities be

changed through implementing the ITD in a

especially easy way, so that the need for skill

acquisition is reduced? An interdisciplinary

approach has been taken to start answering some of

these questions, and to give a first set of directions

so that the possibilities the robot offers can be full

taken advantage of. The following solution approach

gives necessary input to theoretically answer these

questions; the performed experiment gives the first

set of answers.

2 PROBLEM STATEMENT AND

SOLUTIONS APPROACH

2.1 Skill Acquisition Theory

(Ackerman, 1988)

Ackerman distinguishes between three phases within

the skill acquisition process, each of which is

determined by other abilities (Fig. 2). The first

phase, which is referred to as the cognitive phase, is

characterized through slow performance and few

errors. While practicing the learners need to build

productions, which are “if...then...clauses” that

connect a condition specifying when the actions

must be applied and the action itself (Anderson,

1980). If for example the traffic light turns red, then

the car driver must break. The process of building

productions requires both cognitive and attentional

resources (Ackerman, 1988).

Figure 2: Demonstrates the three phases of the skill

acquisition process. The first phase is given via the

straight line, the second one via the dashed and the third

one via the dotted line (adapted from Ackerman, 1988).

After having acquired a broad cognitive

representation of the task, the learner proceeds to the

associative phase. Practicing now results in finding

easier ways to reach the same result, in generalizing

SKILL ACQUISITION PROCESS OF A ROBOT-BASED AND A TRADITIONAL SPINE SURGERY

the productions to other similar tasks, in eliminating

sub-goals and in strengthening the productions. The

ability, which determines this phase’s performance,

is the ability to compare the stored information with

the information about the new situation and to act

based on the outcome of this comparison. These are

defined as psychosensoric abilities (Ackerman,

1988).

In the last phase, which is called autonomous

processing, learners thoroughly understand the task

and have developed efficient productions that can be

performed automatically without or with only few

attentional/cognitive demands. Performing the task

is thus fast and accurate and the performance is only

limited by psychomotor abilities (Ackerman, 1988).

2.2 Factors influencing the

prototypical skill acquisition

process

This description of a skill acquisition process is only

prototypical. As already mentioned cognitive

abilities for example influence the way to learn, so

that interindividual differences result in qualitatively

different skill acquisition processes. However, not

only characteristics of the learner but also the ones

of the task influence the way to learn.

2.2.1 Characteristics of the task

A more complex task is – generally speaking – a

more difficult one, which results in a prolonged

cognitive and associative phase, so that the overall

skill acquisition process takes longer. Regarding the

robot-based spine surgery this has two major

consequences: first, the learning process is shortened

if its productions are less complex and second the

learning process can further be simplified if the

tool’s application is made less complex. A

comparison between the traditional way to perform

surgery and the robot-based one reveals that some

components are added to the traditional way, which

make it more difficult. The surgeon for example has

to “explain” the ITD on a 3D model of the patient,

where the hole should be ideally placed. However,

very difficult components are no longer needed.

When operating with the ITD the surgeon no longer

needs to consider the drill’s resistance in order to

figure out whether he or she is drilling at the right

spot or not. This is a great relief for the surgeon. An

exact and enclosing comparison of both methods is

out of the scope of this article, however it reveals

that the robot-based surgery is altogether less

complex, so that it can be inferred that the skill

acquisition process is shortened.

Task consistency refers to the number of

invariant rules for completing the task in question

successfully (Ackerman, 1987). If the task is

inconsistent, no general valid productions can be

generated, so that the task cannot be acquired. As the

ITD is used in a critical medical context, the rules

for deciding when a given production is to be used

must be highly consistent, so that learning takes

place.

2.2.2 Characteristics of the learner

Interindividual differences do determine

performance during the skill acquisition process:

cognitive abilities influencing the first,

psychosensoric the second and psychomotor abilities

the third phase of the skill acquisition process. The

first two refer to intelligence. The model of

intelligence on which the study is based on is the

Berlin Intelligence Structure Model developed by

Jäger (1982). This hierarchical model distinguishes

between two facets: operations and contents. The

contents refer to the type of material that must be

processed; the operations define what must be done

with the content. The figural content determines

performance during surgery. It for example requires

the surgeons to interpret the information on two-

dimensional X-rays three-dimensionally. An

operation is for example reasoning, which refers to

the ability to solve problems. Perceptual speed,

another operation, is related to the speed of the work

on simple, cognitive tasks. As this description

shows, the latter operation is closely related to the

cognitive requirements in the second phase,

reasoning to the requirements in the first phase.

How does the skill acquisition process change

when comparing two learners, one with high

reasoning, figural abilities, and perceptual speed,

and the other one with low abilities? Learners who

are more able show greater performance levels and

thus proceed to the next phase faster, because they

build productions that are more exact faster. As in

the next phase, this ability no longer predicts

performance and interindividual performance

differences are then balanced, if the learners have

the same level of the ability influencing performance

in the next phase. However, the advantage, the more

able learners have already gained, will not be caught

up by the others, so that learners with higher ability

levels have a general shorter and steeper learning

curve.

The individual differences in both the

performance level and the time needed to reach the

level where the next phase starts can be reduced if

the abilities underlying the first and second phase

can be made less important. As already described, an

ICINCO 2004 - ROBOTICS AND AUTOMATION

increased task complexity makes both the cognitive

abilities and the psychosensoric ones more

important, so that their influence can be reduced

through reducing task complexity. Therefore, the

robot-based and the traditional spine surgery have

already been compared with the result that the robot-

based one has been rated as less complex. Thus, the

first major accomplishment of implementing the

robot is a shortened skill acquisition process.

Another major factor heavily influencing the

skill acquisition are psychomotor abilities.

Psychomotor ability reflects the ability to react fast

on tasks that require little or no cognitive processing

at all (Ackerman, 1988). The abilities influencing

the success of the surgery in question are limited to

arm and finger movements: precision and co-

ordination. A participant with low abilities shows a

slower course of the acquisition process but also

fails to reach the same performance level. This

difference has huge consequences for the success of

a surgery: a surgeon with higher psychomotor

abilities will manage the challenges of the surgery a

lot better than one with less psychomotor abilities

even after having acquired the skill. This is

especially important, as here the drill’s resistance

must be analyzed and the operator must react with

very fine motor movements. Psychomotor abilities

thus decide on success or failure in a traditional

surgery. Here, the second accomplishment of

implementing the robot is apparent: the robot

reduces the impact of different ability levels and

thus enables not only surgeons with very high

psychomotor abilities to perform this type of

surgery.

2.3 Problem Statement

Summarizing, the ITD’s implementation

theoretically has two major accomplishments: it

reduces the impact of psychomotor abilities, so that

at the end of the learning process learners with less

abilities do not show a lower level of performance.

Second it reduces the impact of the cognitive

abilities because of its decreased complexity, so that

the learning process is shorter. However, the positive

effects can be reduced if the robot-based surgery is

made more complex than it needs to be and thus

increases the cognitive workload. Therefore, a

thorough analysis is necessary even before the

completion of the tool to make the robot benefit

from synergies between engineering and

psychology.

The registration of a vertebra, i.e. the matching

of the robot and the patient coordinate frames, is a

good example to demonstrate a possible reduction of

the ITD’ advantages. There are a couple of

algorithms which could work, however these vary

regarding the challenges for the surgeon. One

algorithm with which the matching could be

performed is the surface matching. Here, the surgeon

needs to scan the vertebrae using a pointer. The

scanned area should be as big as possible to ensure

successful matching. The computer tries to align

these scanned points on the patient’s bones with the

3D model to identify the vertebra that must be

performed surgery on. In contrast to the surface

matching, pair-point matching determines exactly

which points of the vertebrae must be scanned. The

surgeon then has to exactly identify the relevant

points on the screen, he or she has to find these

points on the patient and scan them. The probability

of successfully matching the 3D model and the

actual vertebra is here augmented, whereas the

scanning process is more complicated. Theoretically,

the surface matching should be chosen for the ITD’s

implementation. Otherwise the more complex

production needed to acquire the pair-point

matching which would make the robot-based

surgery more complex and thus reduce its positive

effect compared to the traditional surgery.

The experiment was designed in order to test the

impact of the matching procedure on the work load

of the surgeon, but it is also used to start research in

this field, showing that a learning curve is apparent

to validate Ackerman’s theory and implement it as a

basis for future research. These are the research

questions to be addressed:

-Does learning take place when operating the

spine?

-Is the learning curve concordant with

Ackerman’s skill acquisition theory? The theory is

tested empirically via the factors reasoning, figural

abilities, general intelligence, which should have

high predictive validity coefficients for the first

practice trials. The second and third phases was not

tested.

-Do the learning curves differ regarding the

cognitive workload? Altogether two learning curves

result out of this experiment: one without matching,

and one with the pair-point matching. This

hypothesis tests the impact of the performance of the

matching procedure on the skill acquisition process

regarding the cognitive phase. Through adding one

component of the robot-based surgery to the

traditional one, the answer of this hypothesis is an

important step to make the ITD cognitively as easy

as possible for the surgeon and thus to make major

advantage of the ITD’s implementation.

SKILL ACQUISITION PROCESS OF A ROBOT-BASED AND A TRADITIONAL SPINE SURGERY

3 EXPERIMENT

3.1 Participants

The experiment’s participating group was one of

convenience: the participants were recruited in

lectures at the University of Mannheim, so that all

participants were students (50% male, 50% female).

Surgeons could not be recruited due to their strict

time schedule. Also, not enough students of

medicine could be convinced to participate, so that

17 participants study a major related to computer

science, 17 psychology, 5 medicine and 5 were from

other majors such as business. The group size of 44

was set so that possible medium effects can be

detected according to the standards set by Cohen

(1992). The reported average drilling experience was

little, whereas test theoretical problems might have

biased the results.

3.2 Apparatus and instruments used

Reasoning, general intelligence and figural abilities

were measured to predict performance regarding

drilling time and accuracy in the cognitive phase of

the skill acquisition process.

The diagnostic tool used to collect information

on intelligence was the Berlin Intelligence Structure

Test (BIS-4) (Jäger, Süß & Beauducel, 1997). The

short version of the test was used to assess general

intelligence and reasoning. For collecting data on

figural abilities, some further tasks were added to the

test. These resulting 25 tasks were separated into

two booklets, in between which a break of 10

minutes was included. Both the standardization of

the administration and of the analysis given in the

test’s manual were used to ensure objectivity. Other

major criteria of the test were repeatedly measured

for the BIS-4 and showed good results (for a

summary see (Jäger, Süß & Beauducel, 1997)). It

was decided to work with this test first because the

BIS-4 measures all relevant information needed to

investigate the cognitive phase, and second because

the test’s quality criteria are highly promising.

The time needed to drill was measured with a

standard, digital stopwatch during the drilling

processes. The stopwatch was started when the drill

first touched the vertebra and was stopped as soon as

the participant told the experimenter that he or she is

finished. The recorded time was reported to the

participants as feedback, which is necessary to

enhance the skill acquisition process. Regarding

accuracy, the participants were told to drill exactly

20 mm deep, directly in the vertebra’ center, at an

angle of 90

and were told to watch that the surface

area of the drilling to make sure it was not getting

too big. The depth, the surface area, and the

deviation from the center were all based on

measures of an electronic digital caliper

(repeatability accuracy: 0.01 mm). The angle was

assessed with a stick of wood that had the same

diameter as the drill (4mm), which was put into the

drilled hole and was adjusted to a vertical line. With

an angle made of steel the biggest deviation from the

ideal angle of 90

was recorded. If the participants

performed the matching, another variable was

included into the accuracy index: the success of the

matching trials. Feedback on each point was given

after each trial to the participant.

3.3 Experimental set-up

The experiment took place in two sessions. In the

first one, which was between 1.5 and 2 hours long,

participants performed the selected tasks of the BIS-

4. They were tested in groups with a maximum size

of 9 participants. The second session took place

between 1 and 2 weeks later with one participant

each and also took about 1.5 and 2 hours depending

on the condition into which the participants were

grouped randomly: the matching versus no matching

group. The second session started with a short

explanation of the experiment. While reading the

instructions to the participants, the procedure was

explained on a sketch of the “spine” used in this

experiment. These sketches roughly showed the

experimental setting: a piece of wood lying on the

table represented the spine, into which six round

sticks were inserted roundly shaped at the top. These

six pedicles were hidden under a towel during the

whole experiment, so that the participants had no

visual information about their shape, and were used

in order to perform the matching. The participants

had to find the highest point of a pedicle of their

choice and had to touch it with a pointer and then

press a switch lying on the floor. This pointer was

adapted in shape and size to the one used in the

medical setting. The pointer and the pedicles were

electrically connected to a display box. As soon as

the participant pressed the switch, two lights

indicated to the experimenter that first the switch

had been pressed, and second whether the highest

point had been touched or not. The participants

could choose the pedicle they wanted to match with

one constraint: it was not allowed to use the same

one twice in a row.

After having performed two of these matching

trials, one drilling procedure was performed. The

drilling process was the same in both groups: The

hole had to be drilled into similar vertebrae, that

were arranged on the other end of the “spine”: from

ICINCO 2004 - ROBOTICS AND AUTOMATION

16 vertebrae 10 had to be chosen. Each vertebra was

about 30 mm long and 12 mm wide and was inserted

into a bush made of aluminum that was inserted into

the “spine”. The drilling itself was made with a

portable drilling machine. Except the pointer and the

drilling machine, no other tools were used. The

pedicles into which the hole had to be drilled were

not hidden and were straight at the top. Last, a final

questionnaire was answered in order to collect

demographic data on the participants. The spine on

which both the drilling and the matching have been

performed is given in Fig. 3.

Figure 3: The “spine” used in this experiment. On the left

side the vertebrae used for the matching procedure are

displayed, the ones on the right side were used for the

drilling procedure.

3.4 Results

3.4.1 Pre-Analysis

Each variable indicating accuracy was tested with

the F-distribution to figure out whether a learning

effect took place. These are the results:

-For the deviation from the ideal depth the

learning factor is significant with a probability of

p = 0.000 (F(9,387) = 16.687). A learning effect

with an effect size of f

= 0.388 was found indicating

how many standard deviations the participants

improved their drilling.

-For the deviation from the ideal angle the

learning factor, is significant with an error

probability of p = 0.031 (F(9,387) = 2.069). The

effect size was calculated as f

= 0.049.

-For the deviation from the center, no learning

effect took place, which might be based on the

inaccurate feedback given to the participants. The

probability is p = 0.171 (F(9,387) = 1.436), which is

bigger than the accepted critical value of p = 0.05 to

mistakenly reject the hypothesis, that no effect is

there, although one exists. The size of the learning

effect was calculated as f

= 0.033, which shows that

an effect might be there, however, not big enough to

be detectable with the number of participants tested.

-For the surface area the learning factor is

significant with a probability of p = 0.016

(F(9,387) = 2.303). The effect size was calculated as

= 0.053.

-For the matching procedures the learning factor

is significant with a probability of p = 0.000

(F = 9,189) = 4.548. The effect size was calculated

as f

= 0.216.

As the learning curve is analyzed, the accuracy

index is only made up of those variables showing a

learning effect. These variables were transformed to

delete the information about the measurement unit

and about the ideal value, so that each single

variable gives information about the number of units

the ideal value has not been met. The mean was

calculated out of these transformations to indicate

the overall accuracy. The data set was also checked

for extreme values and two participants had to be

excluded.

The drilling time measures showed many

extreme values. As this might result in non-

significant results when testing the hypotheses, the

time measures were transformed logarithmically.

Despite this transformation, one participant had to

be excluded from the analysis.

3.4.2 Significance testing

Finally, the research questions can be answered:

-Does learning take place when operating the

spine regarding drilling time? The analysis shows a

significant learning factor (F(9,360) = 10.661,

p = 0.000) with an effect size of f

= 0.266.

Curvilinearity is tested as well with significant

results: the linear trend is significant with p = 0.000

(F(1,40 = 29.165, f

= 0.729), the quadratic one with

p = 0.000 (F(1,40) = 7.825, f

= 0.679) and the cubic

effect with p = 0.001 (F(1,40) = 12.343, f

= 0.308)

giving information about the learning curve’s shape.

-Does learning take place when operating the spine

regarding accuracy? The learning factor is

significant as well with p = 0.000 (F(9,351) = 0.725,

= 0.566). The linear trend is also significant with

p = 0.000 (F(1,39) = 67.372, f

= 1.713), the

quadratic one with p = 0.000 (F(1,39) = 17.042,

= 0.449) and the cubic one with p = 0.032

(F(1,39) = 4.917, f

= 0.116). Fig. 4 shows the

learning curve of all participants over all trials.

-Is the learning curve concordant with

Ackerman’s skill acquisition theory regarding

drilling time? Testing this interaction effect between

learning and the cognitive factors reveals the

following results: The interaction effect between the

learning factor and reasoning is significant with

p = 0.010 (F(9,333) = 2.446, f

= 0.066), the

interaction effect between the learning factor and

general intelligence is not significant with p = 0.669

(F(9,333) = 0.744, f

= 0.020) and the interaction

SKILL ACQUISITION PROCESS OF A ROBOT-BASED AND A TRADITIONAL SPINE SURGERY

effect between the learning factor and general

intelligence is not significant with p = 0.913

(F(9,333) = 0.440, f

= 0.012). Reasoning shows a

significant quadratic trend with p = 0.016

(F(1,37) = 6.389, f

= 0.172).

-Is the learning curve concordant with

Ackerman’s skill acquisition theory regarding

accuracy? The interaction effect between general

intelligence and the learning factor is significant

with p = 0.010 (F(9,324) = 2.463, f

= 0.069), the

interaction effect between the figural abilities and

the learning factor is significant with p = 0.002

(F(9,324) = 3.003, f

= 0.084) and the interaction

effect between reasoning and the learning factor is

not significant with p = 0.113 (F(9,324) = 1.539,

= 0.042). The curvilinear trends also reveals

significant results: the linear trend is significant for

the interaction between general intelligence and the

learning factor with p = 0.040 (F(1,36) = 4.539,

= 0.131) as well as the linear trend for the

interaction between figural abilities and the learning

factor with p = 0.007 (F(1,36) = 8.238, f

= 0.229).

The cubic trend is also significant but only regarding

the interaction between the learning factor and

general intelligence with p = 0.010 (F(1,36) = 7.361,

= 0.200). Fig. 5 visualizes the different learning

curves for the more and less intelligent participants.

-Do the accuracy’s learning curves differ

regarding the cognitive workload? A three-way

interaction effect between the cognitive factors, the

variable that groups the participants into their

experimental condition and the learning factor must

be tested: the result is not significant for general

intelligence with p = 0.921 (F(9,324) = 0.427,

= 0.010), for reasoning with p = 0.766

(F(9,324) = 0.636, f

= 0.017) and for figural

abilities with p = 0.988 (F(9,324) = 0.245,

= 0.007).

For all tests, the assumptions underlying the

statistical significance tests were checked, and none

of them was severely violated. Further, the stability

of the results was tested revealing very little

variation regarding the reported effect sizes, so that

the results can be interpreted.

3.5 Discussion, conclusion, and final

remarks

First of all, this study shows that it is both necessary

and possible to learn to perform surgery. Drilling

time shows a large learning effect: the participants

got faster with the number of trials performed until

about the 7th trial. It is surprising that learning

interacts with the participants’ reasoning abilities in

the following way: the more able participants need

more time to drill, show a longer learning process

and also improve their timing less than do the less

able participants. It is in this respect concordant to

the theory that reasoning actually predicts the

acquisition process, however, the effect’s direction

must be discussed. Probably the more able

participants focus on accuracy, so that the drilling

time is a less important factor for them. The

participants with greater figural abilities drill more

accurately, need less time to reach a comparable

accuracy index, and acquire the skill faster. The

same is true for general intelligence. As the latter is

based on various components of less general

intelligence factors such as reasoning or figural

abilities, this is not surprising. Especially general

intelligence shows relatively high correlations with

reasoning, however, without causing statistical

problems. This might be a reason for the results

regarding intelligence and drilling time. Altogether,

the results confirm Ackerman’s theory: skill

acquisition takes place for both time and accuracy

and the learning process depends in its first phase on

intelligence. The learning curve for accuracy is

altogether longer than the one for the drilling time.

Figure 4: Line plot of the accuracy levels over all

trials. Smaller numbers indicate better accuracy. The

asymptote shows that the participants do not reach a

perfect accuracy level.

Figure 5: Line plot of the two learning curves for the less

able participants (straight line) and the more able ones

(

dotted line

)

. Smaller numbers indicate better accurac

ICINCO 2004 - ROBOTICS AND AUTOMATION

Here about 9 to 10 trials are needed to reach the

phase in which the cognitive factors no longer play

an important role. Fig. 4 further demonstrated that

human beings do not reach a perfect accuracy level,

and gave a benchmark on the accuracy level of the

human beings that the robot must exceed. As this

“spine surgery” has only been a rough sketch of the

real surgery, one can be imagine how much bigger

the already detected medium-sized effects could be

if tested in the real word. Last, the impact of the

pair-point matching procedure on the skill

acquisition’s cognitive phase is not significant, so

that a possible effect has either been too small to be

detectable with the number of participants tested or

the effect might not be there. The results indicate

that the implementation of the pair-point matching

procedure, thus, does not impede the robot’s positive

impact.

Our simulation demonstrated that the input from

psychology paid off. This article provides methods

to analyze the demands on any technical system in

respect to the human skill acquisition process as an

important system design factor. Skill acquisition

theory worked in predicting and explaining

performance. Therefore, let us look at the demands

to be put on an excellent robot through the lens of

skill acquisition theory:

a) A robot should at least in the long run reduce the

cognitive load on the surgeon. The surgeon then can

invest his or her cognitive resources on other parts of

the surgery.

b) A robot should reduce the time it takes to acquire

a skill. In the long run, this should contribute to

reduce the inevitable costs of introducing it.

c) Reducing cognitive load the robot should enable a

larger percentage of medical doctors to learn and do

high-quality surgery.

d) A robot should reach a level of accuracy sooner

and outperform the benchmarks set by traditional

methods in terms of much higher accuracy.

Will our ITD fulfill these promises? Well, we are

working hard on it and future research will tell you

how much we succeeded.

REFERENCES

Ackerman, P.L., 1987. Individual differences in skill

learning: an integration of psychometric and

information processing perspectives. Psychological

Bulletin, 102(1), 3-27.

Ackerman, P.L., 1988. Determinants of individual

differences during skill acquisition: cognitive abilities

and information processing. Journal of Experimental

Psychology: General, 117(3), 288-318.

Anderson, J.R., 1980. Cognitive psychology and its

implications. New York: W.H. Freeman and

Company.

Cohen, J., 1992. A Power Primer. Psychological Bulletin,

112 (1), 155-159.

Jäger, A.O., 1982. Mehrmodale Klassifikation von

Intelligenzleistungen: Experimentell kontrollierte

Weiterentwicklung eines deskriptiven

Intelligenzstrukturmodells. Diagnostica, 28(3), 195-

225.

Jäger, A.O., Süß, H.-M. & Bedauducel, A., 1997. Berliner

Intelligenzstruktur-Test. Göttingen: Hogrefe.

Pott, P.P., Schwarz, M.L., Köpfle, A., Schill, M, Wagner,

A., Badreddin, E., Männer, R., Weiser, P., Scharf, H.-

P., 2003. ITD - a handheld manipulator for medical

applications: concept and design, 3rd annual meeting

of CAOS, Marbella, Spain.

Siebert, W., 2000. Navigation in der Wirbelsäulenchirur-

gie. Implant, 2, 9-10.

SKILL ACQUISITION PROCESS OF A ROBOT-BASED AND A TRADITIONAL SPINE SURGERY