Content uploaded by Mark Neerincx
Author content
All content in this area was uploaded by Mark Neerincx
Content may be subject to copyright.
MICRO-SCALE SOCIAL NETWORK ANALYSIS FOR ULTRA-LONG SPACE FLIGHTS
D. Hennes1, K. P. Tuyls1, M. A. Neerincx2, and G. W. M. Rauterberg1
1Eindhoven University of Technology, P.O. Box 513, 5600 MB Eindhoven, The Netherlands
2Delft University of Technology, P.O. Box 5031, 2600 GA Delft, The Netherlands
ABSTRACT
This paper proposes to leverage the mathematical means
of game theory to analyze on-board social crew dynam-
ics. We describe how game theory facilitates capturing
the essence of interactive decision making, thereby rais-
ing the potential for a fully automated and unintrusive
monitoring and diagnosis tool. Finally, we present pre-
liminary findings based on the base-line data collection
and the first phase of the ground based Mars-500 isola-
tion experiment.
Key words: Game theory, social network analysis,
Mars 500.
1. INTRODUCTION
A manned mission to Mars is considered to be one of
the ultimate endeavors of mankind. Though not feasi-
ble with state-of-the-art technology, various space agen-
cies are exploring plans to make a human mission to
mars a viable venture. Recent proposals anticipate launch
dates as early as 2035. Besides highly reliable technical
equipment, crew members on their mental and physical
peak are of utmost importance in order to assure success.
The isolated space environment during ultra-long space
flights affects a number of physiological, psychosocial
and mental processes critically involved in human per-
formance. It is vital to a mission’s success to understand
these psychological limits. Past experiences have shown
that the mental health of the crew can have a great ef-
fect on the success or failure of a mission (e.g. long-
term isolation can lead to sleep deprivation, depression,
irritability, anxiety, impaired cognition, and even hostil-
ity). Latent and overt stress factors are mental strain,
lack of capability to rescue crew members, isolation,
monotony, tedium of life aboard an autonomous shuttle,
and interpersonal problems. These issues develop slowly
over time and are very difficult to detect and remedy
for observers on the ground. Computer-interactive inter-
vention programs show promise for adaptation to long-
duration space flights. Software-based delivered pre-
vention and intervention information may be more com-
fortable for crew members than disclosing highly per-
sonal information to others [7, 8]. Programs applying
cognitive-behavioral and self-help instructions show the
potential to be as effective as face-to-face intervention if
dealing with mild to moderate depression, anxiety, and
other types of psychopathology [2, 6, 7, 11]. In order to
trigger the use of such countermeasures effectively one
needs to assess the situation onboard continuously, thus
constituting the need for an automated tool to measure
the individual mental capacity as well as interpersonal
dynamics.
Interpersonal relationships can be captured and analyzed
using the representation of a social network; persons are
represented by nodes while links between nodes describe
or label the relationship features. Considering the small
number of crew members during ultra-long space flights,
we speak of ”micro-scale networks”. In order to deter-
mine the characteristics of links (e.g. cooperation or fair-
ness) we make use of a strategic game, called Colored
Trails - a three-player negotiation game [3]. It is played
on a board of 4 times 4 squares, colored in one of five col-
ors. Each player possesses a piece located on the board
and a set of colored chips. A colored chip can be used
to advance a player’s piece to an adjacent square of the
chip’s color. The general goal is to position pieces onto
or as close as possible to a common goal location. Col-
ored Trails has two distinct roles: two proposers and one
responder. Proposers can suggest a chip exchange to the
responder. The responder can accept exactly one - or no
proposal at all. All players are aware of the board state;
proposers only have information about their own chip set
and the one of the responder; the responder can see all
chip sets. A single game of Colored Trails is played as
a one-shot game, meaning that proposers can only make
a single proposal and the responder can only accept or
reject proposals and is not allowed to make a counter of-
fer. Furthermore, players do not actually have to move
their pieces step-by-step toward the goal location. Once
the responder has reacted to the proposals (by either ac-
cepting one or rejecting both) the chips are exchanged
according to the winning proposal, or stay fixed if the re-
sponder rejected both proposals. With the post-exchange
chip sets to their disposal, players individually advance
on a ”colored trail” toward an optimal location and score
accordingly.
In this work, we use the mathematical foundation of game
theory and evolutionary game theory (EGT) to model and
___________________________________________________________________________
Proc. ‘IJCAI–09 Workshop on Artificial Intelligence in Space’, Pasadena, California, US
17–18 July 2009 (ESA SP-673, September 2009)
Figure 1. Illustration of the Mars-500 facilities (Image credits: IBMP).
study the crew member interactions in the game of Col-
ored Trails. Traditional game theory mainly analyzes ra-
tional decision making, while EGT emphasizes adapta-
tion and learning. Both the ability to think rationally and
to adapt to changing situations are deciding factors for
mission success.
The experiment presented here is part of the Mars-500
study to be carried out at the Institute for Biomedi-
cal Problems (IBMP) in Moscow. The European Space
Agency and the Russian Academy of Sciences jointly
plan and conduct this study in order to simulate a manned
mission to Mars. It provides the unique opportunity to
study crew member interactions while collecting data
about the human subjects’ health and performance dur-
ing experimental isolation. The confinement study imi-
tates all key peculiarities expected to be present during
future missions to Mars (i.e. ultra-long duration, need for
autonomy, affected communication due to signal delay,
and limited stock of expendables). A pretest of 105 days
has recently launched; the full scale experiment with a
duration of 520 days will start in the last quarter of 2009.
The rest of this article is organized as follows. Section 2
provides background information about Mars 500, the
game theoretical framework as well as Colored Trails. In
Section 3 we discuss related work. Section 4 outlines the
methodology used to analyze data collected during iso-
lation and Section 5 presents preliminary findings. Sec-
tion 6 concludes this article with discussion and plans for
future work.
2. BACKGROUND
This section addresses the specifics of the Mars 500 pro-
gram in greater detail (Section 2.1). Besides general set-
ting and operational conditions of Mars-500, this section
also provides the particular prerequisites of the here pre-
sented research. Therefore, we summarize required back-
ground knowledge from the fields of game theory and
evolutionary game theory by means of a simplified ex-
ample (Section 2.2) and hereafter consider more complex
interactions in the game of Colored Trails (Section 2.3).
2.1. Mars-500
In 2004 the Institute for Biomedical Problems (IBMP) in
Moscow and the European Space Agency have started to
plan a full-scale ground based simulation of a manned
mission to Mars. Such a full scale mission requires be-
tween 520 to 700 days of isolation. Referring to the lower
end of this time frame the initiative was named Mars 500.
Ground based isolation studies, including bed rest studies
as well as human missions in low Earth orbit provide cru-
cial experience and isights to evaluate the feasibility of a
manned Martian mission and to list general requirements
for long duration space flights. Human missions beyond
the Earth orbit are affected by radiation hazards, the im-
pact of microgravity on human physiology and various
medical and psychological issues. While the former two
rely on research onboard the International Space Station
(ISS) and future missions to Moon, the affect of long term
isolation can be effectively addressed by ground based
studies such as the Mars 500 program.
The goal of the Mars 500 study is to gather data, knowl-
edge and expertise required to prepare a real mission to
Mars. Hence, all key peculiarities expected to be present
during future missions to Mars are reflected:
•ultra-long duration
•need for autonomy
•affected communication due to signal delay
•and limited stock of expendables.
This ensures that psychological and physiological im-
pacts of isolation through such an extended period of time
are observed as close to reality as possible.
A crew of six candidates (four Russians and two Euro-
peans) are sealed insight the facilities of the Institute for
Biomedical Problems in Moscow. This crew is chosen
to aggregate a very diverse field of expertise and skills,
including knowledge in the field of aviation, biology, en-
gineering, medicine and physiology. A initial 105-day
isolation period has recently started in March 2009 and
is currently underway. Once completed successfully, a
full 520-day study is planed to launch in the first quar-
ter of 2010 in which all elements of a Martian mission
are being simulated: transit to Mars, orbiting the planet,
landing maneuvers, Mars surface operations and return
to Earth. The facilities at IBMP accommodate for these
different simulation episodes with a number of intercon-
nected isolation chambers. An illustration of this setup
is depicted in Figure 1. Part of the isolation facilities re-
semble the hermetically sealed mock-up space craft: a
habitable module, a utility module and the medical unit.
Furthermore a landing ship and the Martian surface itself
are being simulated.
Throughout the isolation period and all episodes of sim-
ulation, the crew needs to be self-reliant to a great ex-
tend. This includes monitoring life support, control re-
source consumption, maintaining technical equipment as
well as performing a number of scientific experiments in
which the participants act as subjects themselves. In this
work we present an experiment where crew members en-
counter in strategic game interactions while data is being
gathered about their interpersonal dynamics. The next
section describes the mathematical framework of game
theory and evolutionary game theory that allows to quan-
tify these interactions.
2.2. Game theory and evolutionary game theory
Traditional game theory studies strategic interactions
where multiple players act rational in order to maximize
their expected payoffs. These players can for instance
represent companies in the economical area, single indi-
viduals in the psychological or social science domain as
well as computer agents situated in a multi-agent setting.
To provide an intuitive understanding of the methodol-
ogy of game theory we shall first consider a particular
strategic encounter. The Prisoners’ Dilemma is a preva-
lent example to showcase game theoretical modeling and
analysis practice. This chapter first introduces the classi-
cal predicament of two suspects, while gradually extend-
ing the problem to cover iteration, evolution and finally
multi-state settings.
The dilemma presents itself in the following situation.
Two subjects are under arrest. The police cannot provide
enough evidence to bring in a severe accusation. Thus the
two prisoners are separated for interrogation and are both
confronted with the same choice: either confess the crime
and hence defect, or to deny any accuse and hence coop-
erate with the other suspect. If one suspect testifies while
the other remains silent, the defector is released while his
loyal fellow is convicted to serve full sentence. If both
defect, both receive the full sentence. If both deny all
accuses, they serve a short sentence for a minor charge.
How should the prisoners choose rationally in order to
minimize penalty?
Seemingly, the best option is to cooperate, both parties
only serve a short sentence and are set free after a rel-
atively short time. However, this intuition suggests a
non-rational decision. Game theory provides the tools to
reason about such a strategic situation and identify the
rational choice. Let us consider the options of a sus-
pect under the following set of assumptions. A suspect
is self-interested and therefore only cares about his own
well-being (or degree of punishment in that perspective)
and thus does not profit from lowering the sentence of
the other. Furthermore, any communication between sus-
pects is inhibited and both are ensured that the other is
provided with the same information and faces the same
choice. Both have the same option to chose exactly one
of the two actions defect or cooperate. In other words,
all action sets and corresponding payoffs are equivalent,
therefore we call such a game symmetric. Since no com-
munication is allowed we might as well assume that play-
ers act simultaneously. The payoff of each player is not
solely dependent on its own choice but rather on the col-
lective joint action. Suppose a player chooses to coop-
erate. If the other one cooperates as well the player is
convicted to serve a short sentence; if the other one be-
trays him he serves the full sentence. If he instead choses
to defect, he is set free in the former case while as well
serving the full sentence in the later case. We observe that
cooperating is dominated by defecting, i.e. for each ad-
versarial response, defect yields as least as much payoff
as cooperate. Thus players will make the rational deci-
sion to mutually defect and hence end up serving both
the maximum sentence. While intuition suggests other-
wise, in the perspective of rationality the suspects face an
unwinnable situation. This paradox clearly shows the dif-
ficulty in analyzing non-zero-sum situations (i.e. the gain
of one party is not necessarily a loss of another).
We now revisit our initial example, the Prisoners’
Dilemma, in order to bridge the gap between traditional
and evolutionary game theory. The classic dilemma sit-
uation can be extended in different ways - one of them
is to play the game for multiple stages. In the Iterated
Prisoners’ Dilemma the two suspects encounter the same
predicament over and over again. Clearly, this can af-
fect the equilibrium strategies. If the game is repeated in-
finitely often, players can obtain the Pareto efficient equi-
librium reward by invariably cooperating. If however the
game is played Ntimes and this information is known to
the players, the rational strategy is to always defect which
can be shown by backwards induction [5].
In [1] a tournament version of the Iterated Prisoners’
Dilemma is proposed. Each player has to choose a strat-
egy, possibly depending on memory of previous encoun-
ters. Interactions between pairs of individuals occur on a
probabilistic basis while natural selection favors individ-
uals that have performed better in the past. The derived
model as well as results of the tournament show how co-
operation based on reciprocity can evolve in a mechanism
that advocates self-interest. The simplest deterministic
strategy, called Tit-for-Tat, has won the tournament. Tit-
for-Tat starts by cooperating and hereafter responds by
playing the action the opponent has played in the last
stage.
The primary objective of evolutionary game theory is to
model the dynamics of strategy changes in iterated games
such as the Iterated Prisoners’ Dilemma. In particu-
lar, interactions are modeled using biological-inspired ge-
netic operators such as natural selection and mutation. At
each time step two randomly matched individuals from a
population play a particular pure strategy while their pay-
off determines the replication success of the represented
strategies. The process of population change is captured
by the replicator dynamics.
2.3. Colored Trails
While the prisoners dilemma is a good example to intro-
duce traditional game theory and evolutionary game the-
ory it is very limiting and not suitable to capture more
complex inter-social dynamics as present during ultra-
long space flights. For the first iteration of the Mars-500
experiment, the 105-day pretest, we are primarily inter-
ested in games that feature the following properties:
•Simple enough for analysis
•Rich enough to reflect features of
real life interactions
•Grounded in a situated task domain
•Strategic, i.e. partial information to
promote complex reasoning
•Suited to measure social factors
such as fairness
The Colored Trails framework developed at Havard Uni-
versity, School of Engineering and Applied Sciences
serves all these requirements [4]. Specifically, we use
a three-player negotiation variation [3]. It is played on a
board of 4 times 4 squares, colored in one of five colors:
blue, green, red, yellow and grey. Each player possesses
a piece located on the board and a set of colored chips.
A colored chip can be used to move a players piece to
an adjacent square (diagonal movement is not allowed)
of the same color. The general goal is to position pieces
Figure 2. Colored Trails board
onto or as close as possible to a goal location indicated
by a flag. Although, there is a single goal, each player re-
ceives points purely based on its own performance. Col-
ored Trails has three distinct roles: proposer 1 (P1), pro-
poser 2 (P2) and responder (R). Figure 2 shows an exam-
ple of the board, goal and player locations.
Proposers can propose a chip exchange to the responder.
The responder can accept exactly one - or no proposal at
all. All players are aware of the board state; proposers
only have information about their own chip set and the
one of the responder while the responder can see all chip
sets. A single game of Colored Trails is played as a one-
shot game. This means that proposers can only make a
single proposal and the responder can only accept or re-
ject proposals and is not allowed to make a counter offer.
Furthermore, players do not actually have to move their
pieces step-by-step toward the goal location. Once the re-
sponder has reacted to the proposals (by either accepting
one or rejecting both) the chips are exchanged according
to the winning proposal; or stay fixed if the responder re-
jected both proposals. Then, the best possible sequence
of moves is automatically computed and each player re-
ceives a personal score.
Game phases
A game of Colored Trails is divided into a sequence of
three phases and final evaluation. During each phase a
timer indicates how much time is left in the current phase.
Initial phase
In this phase the game board and the chip stacks are pre-
sented to the players. This initial phase allows the player
to locate its own piece on the board and get acquainted
with the chip distribution.
Proposal phase
In this phase the two proposers (i.e. P1 and P2) can make
chip exchange offers to the responder. The Proposers
chips are always located on the left hand side of the win-
dow, the responders chip stack on the right hand side.
Reaction phase
In the reaction phase, the responder is presented with the
two proposals. After evaluating the situation he can either
accept one of the proposals or reject both.
Figure 3. Colored Trails: three player negotiation game. From left to right: view of proposer 1 (P1), proposer 2 (P2) and
responder (R). Note that placeholder figures are exchanged for pictures of the corresponding players.
Evaluation
The final evaluation is fully automatic; players do not
need to act in this phase. The game server automati-
cally computes the best possible sequence of moves for
each player and assigns personal scores. Points are calcu-
lated as follows. For reaching the goal location a player
receives 100 points. If he does not reach the goal, 25
penalty points are subtracted for every square between
the goal and the players position. In addition, for every
chip the player has not used, he receives 10 extra points.
3. RELATED WORK
This section presents a selection of related work. In par-
ticular, we refer to the bioengineering system Homeostat
(Section 3.1) and the analyses of complex strategic in-
teractions using evolutionary game theory as proposed
by Walsh et al. [13] (Section 3.2).
3.1. Homeostat
Homeostat is one of the experiments participating in the
Mars 500 study. The ”Homeostat apparatus” is a system
of interconnected devices, each operated by one of the
subjects. Each device features an indicator and a control
knob; rotating the knob influences the position of the in-
dicator. The indicator is not directly linked to the control
knob of the corresponding unit but rather responds to the
operation of other devices as well. Every subject is in-
structed to fix the position of the indicator to a specific
position. Due to the interaction between control loops,
actions of each individual mutually induce interference.
The collective group problem can only be solved through
well coordinated actions. The selection of interaction
structures (e.g. tree, chain or cycle) and interaction co-
efficients determine the degree of mutual influences and
allow for a variety of experimental setups.
The Homeostat system has been previously used during
a number of isolation and confinement experiments, in-
cluding the 60-day ESA EXEMSI campaign [9], the 135-
day isolation during the ”Human Behaviour in Extended
Spaceflight” (HUBES) study in the Mir orbital station
mock-up at IBMP [12] and the ”Simulation of the Flight
of the International Crew on Space Station” (SFINCSS).
Homeostat has proven to be suitable for on-line inferring
of interaction effectiveness of an isolated crew with com-
plex communication structure. Furthermore, leadership
and ”follower” tendency can be detected; the internal bal-
ance between these complementary behaviors is a decid-
ing factor for the effectiveness of the group.
3.2. Evolutionary game theoretic analysis of com-
plex games
In Section 2.2 we have introduced the notion of expected
payoffs to quantify the success of actions taken during
strategic interactions between players. Once these strate-
gic encounters become more complex (i.e. increased
number of players or actions), traditional game theoretic
analysis becomes intractable.Walsh et al. [13] propose to
use heuristic strategies rather than atomic actions and to
compute a so called heuristic payoff table for these prim-
itives. A heuristic payoff table captures the average pay-
off of each heuristic strategy for all possible mixtures of
strategies for a finite number of players. This approach
is suitable for games where the underlying rules are well-
specified and common knowledge though they may de-
scribe very complex, repeated interactions between mul-
tiple agents. Rules specify particular actions agents may
take depending on the the state of the game. Each agent
follows a specific strategy selected from the set of heuris-
tic strategies. Hence, the heuristic payoff table consti-
tutes an abstract representation of the underlying game
with the focus on heuristic strategies rather than atomic
actions. Utilizing this methodology allows to apply for
even more complex encounters the same game theoretic
analysis as for simple games, e.g. the prisoners’ dilemma.
Walsh et al. [13] apply this approach to the auction do-
main and analyze a simulated continuous double auction
with predefined trading strategies as heuristics.
Heuristic payoff tables have been proven to be benefi-
cial beyond the domain of auctions and trading strategies.
In [10] the same approach has been used to analyze the
game of Poker (No-Limit Texas Holdem) using empirical
data collected during human play. The authors use ex-
pert domain knowledge found in Poker literature to clus-
ter complex atomic actions into a small set of heuristics,
called meta strategies. This allows to construct a heuristic
payoff table and perform an evolutionary game theoretic
analysis; which enables to discover switching behavior
between meta strategies.
4. METHOD
In this section we concisely explain the methodology we
propose to perform the evolutionary game theoretic anal-
ysis of the data collected during Mars-500.
1. Collecting interaction data
The game Colored Trails serves as a strategic, interactive
and situated task domain. We record full state informa-
tion, proposals, responses as well as response times for
each action.
2. Clustering atomic actions into meta strategies
Similar to the case of Poker, atomic actions in the game
of Colored Trails are far too complex to analyze directly.
However, clustering atomic actions into a small set of
meta strategies is not straight forwarded either: expert
domain knowledge for this specific game is lacking. Fur-
thermore, the efficiency of automated clustering algo-
rithms strongly depends on an appropriate distance mea-
sure. This distance meassure needs to be defined over (a
set of) features that describe atomic actions within Col-
ored Trails. Both, the possibility to devise ”hand-picked”
rules clustering actions into meta strategies and the de-
velopment of an adequate distance metric are the subject
of current investigations.
3. Computing the heuristic payoff table
The heuristic payoff table represents the payoff table of
the Colored Trails game for the different meta strategies
the subjects can employ during play. It summarizes the
success of a certain meta strategy depending on meta
strategies chooses by other players. Hence, it simplifies
the complexity of the game by using an abstract represen-
tation based on a small set of meta strategies.
4. Hypothesis testing and correlation
Once the heuristic payoff table is computed, evolution-
ary game theory analysis can be used to test different
hypothesis, e.g. (a) the strategic behavior of subjects is
influenced by personal preferences and not only guided
by rational decision making; (b) subjects form coalitions
while excluding other subjects from the group; (c) if the
task becomes more complex, time pressure will effect the
Table 1. Preliminary data collection for the 105-days
study. For both groups the number of session and games
are listed as well as the playing time.
Group Sessions Games Time
pre isolation
Group 1 2 41 40 min
Group 2 2 42 35 min
in isolation
Group 1 2 31 23 min
Group 2 2 29 20 min
total
Group 1 4 72 63 min
Group 2 4 71 55 min
degree of influence induced by personal preferences. Fur-
thermore, it needs to be shown if these findings correlate
with other psychological investigations during Mars-500,
e.g. questionnaire studies and the Homeostat experiment,
as well as information obtained during debriefing.
5. PRELIMINARY RESULTS
The detailed analysis and discussion of data obtained dur-
ing the 105-days study remains for future work. How-
ever, this section augments the general description of
our methodology with preliminary results from the first
weeks of the experiment.
Table 1 lists the number of games played prior to the iso-
lation period as well as during the first weeks of isolation.
Each group played four sessions with a total of 72 and 71
games respectively. In the proceeding analysis we iden-
tify subjects using a group prefix ”1.” or ”2.” coupled
with a label ”A”, ”B”, ”C” to distinguish between the
three subjects within a group.
Clustering of atomic actions in Colored Trails is not
straight forward. Therefore, let us restrict the analysis
to the behavior of the responder. Note that roles rotate
in each group, thus all subjects become the responder at
equal intervals. Table 2 shows the absolute number of
proposals accepted by the different responders as a func-
tion of the proposing individual. Subject 1.A accepted
14 proposals from 1.B, 7 proposals from 1.C and in three
games both proposals got rejected. The chipset distribu-
tion does not favor one of the two proposers and therefore
we can suspect that this imbalance hints at either
•subject 1.B is more generous while proposing, or
•a social preference for 1.B over 1.C.
To test further investigate this hypotheses we compute the
payoffs associated with the meta strategies (1) ”always
1.A
1.B
1.C
115.7
± 18.0
102.5
± 30.1
117.7
± 14.8
117.9
± 17.3
114.3
± 16.9
108.6
± 15.2
2.A
2.B
2.C
88.1
± 32.5
96.0
± 20.9
103.2
± 14.6
120.0
± 10.0
92.1
± 19.8
NA
Figure 4. Illustration of the Colored Trails payoff table for the two groups of players. Arcs are labeled with the average
payoff received by an individual (end of arrow) while accepting a proposal from another individual (origin of arrow).
accepting proposals from 1.B” and (2) ”always accepting
proposals from 1.C”.
Meta strategy (1) yields an average payoff of
115.7±18.0and meta strategy (2) results in an av-
erage of 117.9±17.3. While both strategies award a
comparable payoff, subject 1.A still favors proposals
from 1.B over the proposals from 1.C with a ratio of
2:1. The full payoff tables for both groups is illustrated
in Figure 4. In Group 2 we find a similar imbalance
between proposal acceptance rate and associated average
payoff. Subject 2.A accepted far more proposals from
subject 2.B despite the higher average payoff obtained
by accepting proposals from subject 2.C. An extreme
case of such coalition forming is reflected by the fact that
subject 2.C has not accepted a single offer from subject
2.B.
Once data is recorded over a longer period, payoff val-
ues can be recomputed over different time intervals. This
allows to discover behavioral changes using the evolu-
tionary game theoretic analysis as described in Section 4.
With the aim to detect when coalitions form or brake up
and how crew dynamics evolve over time.
Finally, preliminary data in Figure 4 shows that Group 1
succeeds to cooperate more efficient than Group 2 as in-
dicated by a higher overall payoff flow, to be precise the
weighted average payoff compares 112.9(Group 1) to
96.4(Group 2).
6. DISCUSSION AND CONCLUSIONS
While this paper has only presented results based on pre-
liminary data, it is clear that the crew dynamics during the
play of Colored Trails show interesting coalition form-
ing behavior. This strenghens our hypothesis that play
in Colored Trails is not purely based on rational decision
making but rather influenced by personal preferences. It
has yet to be investigated if these findings correlate with
psychological investigations of various aspects of crew
behavior during Mars-500, e.g. questionnaire studies and
Table 2. Absolute number of proposals accepted by the
different responders as a function of the proposing indi-
vidual (within brakets).
Responder Proposal 1 Proposal 2 Reject
1.A 14 (1.B) 7 (1.C) 3
1.B 7 (1.A) 12 (1.C) 5
1.C 15 (1.A) 7 (1.B) 1
2.A 18 (2.B) 3 (2.C) 1
2.B 7 (2.A) 15 (2.C) 0
2.C 17 (2.A) 0 (2.B) 4
the Homeostat experiment, as well as information ob-
tained during debriefing.
To summarize, we proposed to leverage the mathemati-
cal means of game theory to analyze interpersonal crew
dynamics during ultra-long duration space flights. We de-
scribe how game theory facilitates capturing the essence
of crucial interactive decision making processes, thereby
raising the potential for a fully automated and unintrusive
monitoring and diagnosis tool; such a tool is the prerequi-
site to effectively trigger computer-based self-help coun-
termeasures to depression or anxiety. Finally, we have
presented preliminary findings based on the first 105-day
test phase of the ground-based isolation study Mars-500
and outlined our research agenda for the full 520-day iso-
lation period.
ACKNOWLEDGMENTS
This research was partially funded by the Netherlands
Organization for Scientific Research (NWO) and the
Netherlands Institute for Space Research (SRON).
REFERENCES
[1] Axelrod, R. and Hamilton, W. (1981). The evolution
of cooperation. Science, 211:1390–1396.
[2] Cavanagh, K. and Shapiro, D. (2004). Computer
treatment for common mental health problems. Clini-
cal Psychology, 60:239–251.
[3] Ficici, S. and Pfeffer, A. (2008). Modeling how hu-
mans reason about others with partial information. In
Proceedings of the Seventh International Conference
on Autonomous Agents and Multiagent Systems.
[4] Gal, Y., Grosz, B. J., Kraus, S., Pfeffer, A., and
Shieber, S. (2005). Colored trails: a formalism
for investigating decision-making in strategic environ-
ments. In Proceedings of the IJCAI Workshop on Rea-
soning, Representation, and Learning in Computer
Games.
[5] Gintis, H. (2000). Game Theory Evolving. A
Problem-Centered Introduction to Modelling Strategic
Interaction. Princeton University Press.
[6] Greist, J., Marks, I., Baer, L., Kobak, K., Wenzel, K.,
Hirsch, M., Mantle, J., and Clary, C. (2002). Behavior
therapy for obsessive-compulsive disorder guided by
a computer or by a clinician compared with relaxation
as a control. Clinical Psychiatry, 63:138–145.
[7] Kanas, N. et al. (2006). Psychology and culture dur-
ing long-duration space missions. Technical report, In-
ternational Academy of Astronautics Study Group on
Psychology and Culture During Long-Duration Space
Missions.
[8] Leon, G. (1999). Select-in and countermeasure con-
siderations for long duration crews. In Proceedings of
the 29th International Conference on Environmental
Systems.
[9] Novikov, M., Bystritskaya, A., Eskov, K. N., Vasi-
lyiev, V. K., Vinokhodova, A. G., and Davies, C.
(1993). Homeostat - a bioengineering system. In Pro-
ceedings of the 23rd International Conference on En-
vironmental Systems.
[10] Ponsen, M., Tuyls, K., de Jong, S., Ramon, J., Croo-
nenborghs, T., and Driessens, K. (2008). The dynam-
ics of human behaviour in poker. In Proceedings of the
20th Belgium-Netherlands Conference on Artificial In-
telligence (BNAIC 2008), pages 225–232, University
of Twente, Enschede, The Netherlands.
[11] Proudfoot, J., Goldberg, D., Mann, A., Everitt, B.,
Marks, I., and Gray, J. (2003). Computerized, interac-
tive, multimedia cognitive-behavior program for anxi-
ety and depression in general practice. Psychological
Medicine, 33:217–227.
[12] Savilov, A., Bayevsky, R., Bystrytskaya, A.,
Gushin, V., Manovtsev, G., Nichiporuk, I., Novikov,
M., Ponomareva, I., and V.P.Salnitsky (1997). On-
board equipment-based study of the dynamic of
psycho-physiological and biomedical responses of the
operators during 135-day isolation in the mir orbital
station mock-up. Aerospace and Ecological Medicine,
5:28–38.
[13] Walsh, W., Das, R., Tesauro, G., and Kephart, J.
(2002). Analyzing complex strategic interactions in
multi-agent systems. In Proceedings of the AAAI
Game Theoretic and Decision Theoretic Agents Work-
shop.
