Conference PaperPDF Available

Role of Pattern Recognition in Computer Games

Authors:

Abstract and Figures

We discuss pattern recognition in the context of computer game. The purpose of pattern recognition is to extract relevant information from the game world. This high level information is needed by a decision-making system, which is responsible for producing actions to the game world. We delineate where pattern recognition can be applied in computer games, what are its roles, and what is expected from it. We discuss how pattern recognition can be utilized in different levels of detail, and how the relation between the synthetic player and the human player affects to the requirements for pattern recognition.
Content may be subject to copyright.
Role of Pattern Recognition in Computer Games
Timo Kaukoranta, Jouni Smed
Department of Information Technology and Turku Centre for Computer Science (TUCS),
University of Turku, Lemmink
¨
aisenkatu 14 A, FIN-20520 Turku, Finland
timo.kaukoranta@cs.utu.fi, jouni.smed@cs.utu.fi
Harri Hakonen
Oy L M Ericsson Ab, Telecom R&D, Joukahaisenkatu 1, FIN-20520 Turku, Finland
harri.hakonen@lmf.ericsson.se
Abstract
We discuss pattern recognition in the context of computer
game. The purpose of pattern recognition is to extract rel-
evant information from the game world. This high level
information is needed by a decision-making system, which
is responsible for producing actions to the game world. We
delineate where pattern recognition can be applied in com-
puter games, what are its roles, and what is expected from
it. We discuss how pattern recognition can be utilized in
different levels of detail, and how the relation between the
synthetic player and the human player affects to the re-
quirements for pattern recognition.
1. Introduction
Computer games (CGs) are an unique application area for
pattern recognition. They are the products of their devel-
opers, and, at first sight, it is easy to assume that the de-
velopers know inherently what phenomena occur in their
game world. However, as the worlds in CGs try to resem-
ble more closely the real world, their complexity increases
greatly. Consequently, the need for pattern recognition has
become more prominent. Also, to implement challenging
synthetic opponents computer should recognize the behav-
ior of a human player. In this context, the purpose of pat-
tern recognition is to abstract relevant information from the
game world and to construct concepts and to deduce pat-
terns from this information. These concepts are needed for
higher level reasoning and decision-making. In CGs, this
is associated with artificial intelligence (AI) problems.
There exists a large amount of scientific articles
and books on the fundamentals of pattern recognition
and its applications to several different problem set-
tings. Often pattern recognition is called classification or
identification—or even AI in general. However, pattern
recognition is not one particular technique, but it includes
several techniques such as string and clustering algorithms.
Unfortunately, in the development of AIs for CGs the scope
of pattern recognition has not been widely realized. Pat-
tern recognition could provide essential information for the
decision-making system, like what phenomena and events
have occurred in the observed (dynamic) system or world.
Thus, pattern recognition is only a part—but very impor-
tant one—of whole AI system.
Our motivation is to do pattern recognition ourselves to
discern where pattern recognition can be applied in CGs,
what are its roles, and what is expected from it. For most
parts, we raise questions and provide examples rather than
try to propagate certain techniques or approaches; for ex-
ample, how pattern recognition can be utilized in several
levels of detail, or how the relation between the synthetic
player and the human player affects to the requirements for
pattern recognition. Only after the ground have been laid
out, we can start to think more concretely on the terms of
algorithms. Unfortunately, this falls beyond the scope of
the present paper. Instead, we try to form a coherent pic-
ture of the needs of CGs.
The plan of the paper is following. Introduction to the
concepts of pattern recognition is given in Section 2. Roles
of pattern recognition in CGs are discussed in Section 3.
In Section 4, we point out several topics of CGs related
to pattern recognition. Finally, conclusions are drawn in
Section 5.
2. Fundamentals of Pattern Recognition
Pattern recognition techniques are often lumped together
with AI techniques. However, pattern recognition could
also be seen as a part of whole AI system. Here, we
consider an AI system to be composed of two parts: pat-
tern recognition and decision-making. The task of pattern
recognition is to extract relevant information from the data
source (i.e., to classify the data into classes that identify
Pattern
recognition
Observed
events and states
Decision-making
system
Possible
actions
Requested
actions
World
Primitive events
and states
Previous
primitive events
Figure 1: Relations between the world, pattern recognition,
and decision-making.
interesting phenomena). In some applications this infor-
mation extraction is enough but other AI applications may
need to perform higher level reasoning. This reasoning can
be based on the information extracted by a pattern recogni-
tion technique.
The relation of pattern recognition and decision-making
in an AI system is illustrated in Figure 1. The world, which
can be real or simulated, consists of primitive events and
states (i.e., phenomena) that can be perceived. From these
and previously stored perceptions, pattern recognition pro-
duces higher level observations, which are forwarded to the
decision-making system. The worldallowsa set of possible
actions, and the decision-making system has the responsi-
bility to choose an appropriate action based on the infor-
mation given by pattern recognition.
The simplest form of pattern recognition is pattern
matching, which is a process of searching a predetermined
pattern from given input. If the pattern is a fixed template,
the pattern matching detects the existence of the pattern in
the input (e.g., searching a word from text). Pattern can be
expressed also as a rule, and in this case the matching is an
optimization problem. This optimization bounds the free
variables of the rule (e.g., fitting regression line into a data
set). Generally, pattern recognition can be used into much
more complex problems than pattern matching.
2.1. Concepts
Pattern can be defined as a actual phenomenon of the world
that we want to recognize. For example, pattern can be
an attack or lack of resources. From the data source (i.e.,
the world) we can perceive simple measurements. By as-
signing some primitive meaning to a measurement we have
a feature, which can be numerical (e.g., quantity), sym-
bolic (e.g., name or label), or complex (e.g., combination
of primitives). Feature selection is a process of choosing
the features to be extracted (i.e., relevant data). Extracted
features should be discriminative, descriptive, and compu-
tationally feasible.
Features can also be called as symbols, and a set of sym-
bols forms an alphabet Σ. The selection of symbols de-
pends on the abstraction level of the decision-making sys-
tem. Moreover, in a hierarchical pattern recognition system
the recognized patterns can be symbols for a higher level
pattern recognition. To reduce the computational burden
of the recognition process, the size of an alphabet can be
decreased by quantizing the measurements at the cost of
losing information.
A pattern class is a set of similar patterns. In other
words, the classes should discriminate different types of
patterns. The class set can have a priori knowledge based
on expertise. It can also have a posteriori knowledge when
it is formed in a learning process. In the classification pro-
cess an input data is assigned into at least one class based
on the extracted features. In fact, recognition can be de-
fined as an ability to classify.
2.2. Recognition Methods
Pattern recognition approaches can be categorized into sta-
tistical, syntactic (or structural), and neural pattern recog-
nition [8]. Statistical pattern recognition assumes that ex-
tracted feature vectors obey some probability density func-
tion. Typically, it requires background information from
the application area. Syntactic pattern recognition consid-
ers interrelationships and interconnections of the features
to produce structural information. Neural pattern recogni-
tion approaches try to simulate biological neural systems
by learning process. They can also be seen as a black-box
way to implement statistical and syntactic methods. In-
deed, the boundaries between these approaches are vague
and fading.
In the design and implementation of pattern recognition
system, the task is to absorb significant patterns and rules
from the given data. In other words, the task is to learn
the characteristics of the world. In a supervised learning
process, the designer provides correct identification for all
inputs, and the process constructs the rules from this in-
formation. In an unsupervised learning process, the goal
is to define the classes, which is typically performed by
a clustering method. This process requires an appropriate
training set of features. The variety of actual pattern recog-
nition methods is wide. They utilize, for example, string al-
gorithms, formal languages, clustering techniques, genetic
algorithms, neural networks, and fuzzy logic.
3. Pattern Recognition in Computer Games
Pattern recognition in CGs can be viewed from several
perspectives. In the following we consider first how pat-
tern recognition operates on different levels of decision-
making. Then, we describe the roles of pattern recogni-
tion in a relation between a synthetic player and a human
player. We discuss open game world and story generation,
and show how modeling can be used in both prediction and
production. Finally, we list some common tasks where pat-
tern recognition could be utilized.
3.1. Strategical, Tactical, and Operational Level
By using classical three-level design hierarchy we can
divide decision-making problems into three categories:
strategical, tactical, and operational. Accordingly, the
suitability of pattern recognition methods depends on the
level of the decision-making. Let us now categorize com-
mon problems and features of pattern recognition.
On the strategical level, pattern recognition works over a
long period of time. The amount of data can be large, and,
therefore, the main problem is to filter it down to a suitable
alphabet. Due to quantization, information is lost and the
problem is to ensure that no vital knowledge is left out in
this process. The reason for the high volume of data is that
it is assembled from all inhabitants, items, and events in
the game world. Naturally, we cannot operate with such
large set of data. Decision-making in strategical level is
speculative (i.e., What-If scenarios). This is usually done
offlineor in the background. Therefore, it caninvolve high-
quality pattern recognition methods. Another reason to aim
at good quality is that the cost of a wrong decision is high
on the strategical level.
Tactical level connects the strategical level to the opera-
tional level. Because tactical decisions are made more fre-
quently than strategical, pattern recognition has less time
to use. Consequently, the quality cannot be as high as on
upper level. If the strategy answers what should be done,
tactics describes how to actuate it. Action on the tactical
level considers a group of entities and their cooperation.
Input data for recognition originates from the actors of the
operational level, and, therefore, pattern recognition should
observe the operational environment.
Operational level is concrete and closely associated with
the properties of the game world. Operations are related
to atomatory entities, and, therefore, pattern recognition
techniques must be reactive, short-term, and real-time. Be-
cause the decisions are made for a short term, less accurate
pattern recognition does not necessary lead to irrevocable
problems. For instance, if the chosen path turns out to lead
a dead-end, we can still reroute the path to the destination.
Regardless of the level of decision-making, one should al-
ways bear in mind Patton’s law: A good plan today is better
than a perfect plan tomorrow.
3.2. Enemy, Ally, or Neutral
Although it may seem more natural to think a computer
as an enemy, it may as well be an ally. For example, in
a real-time strategy (RTS) game, player’s interface could
be augmented by implementing a synthetic reconnaissance
officer. It could report on enemy movements and even sug-
gest effective counteractions. In this case pattern recog-
nition has to account human perspective, since the result
will be given to a human player instead of passing it to a
decision-making system.
The third role for the computer player is to act neu-
tral. For instance, in a sports game, like ice hockey or car
racing, a commentator could highlight the events and pro-
vide background information. Obviously, pattern recogni-
tion is needed to recognize what is happening in the game
world. Interest in this problem has increased lately; for ex-
ample, synthetic commentators have been demonstrated in
RoboCup [7].
In addtion to synthetic commentators, the computer
could act as an autonomous camera director. The camera
director programs require sophisticated pattern recognition.
For instance, the placement and angle of the cameras in a
sports game is usually dictated by the television practice,
and the problem is to choose an appropriate camera view
(including the angle and zoom) among them all. If there
is only a single point of interest, like a human-controlled
racing car, the problem is easy; however, if there is action
all around the playing field, it is hard to recognize what to
show and in what order so that the cuts still form a coher-
ent whole. Things become even more complicated, if the
camera can move freely around the game world.
An important neutral role is to act as a referee. How-
ever, some rules may be hard to judge because in order to
detect their state complex decision-making is required. For
example, in soccer the referee can allow the play continue
“when the team against which an offence has been com-
mitted will benefit from such an advantage” and penalize
“the original offence if the anticipated advantage does not
ensue at that time” [3]. To implement this in a computer
game would require pattern recognition which can inter-
pret (possibly complex) causality between the offence and
the subsequent events.
3.3. Open Game World and Story Generation
A game is not a story: while a story progresses linearly,
a game must provide an illusion of free will [2]. Obvi-
ously, the player must have a range of actions to choose
from at each stage. More formally, let us consider the
(a) (b)
a
1
a
2
a
3
a
4
s
i
s
j
s
k
s
l
s
m
a
1
a
2
a
3
a
4
s
i
s
j
s
n
(c) (d)
a
1
a
2
s
i
s
j
s
k
a
1
s
i
s
j
Figure 2: In a game graph, game states are represented as
nodes and actions as directed arches. (a) A linear progres-
sion (e.g., a story) allows no diversion. (b) Outdegree is the
number of arches leaving a node (in this case node s
i
has
an outdegree of 2). (c) Indegree is the number of arches
entering the node (in this case node s
n
has an indegree of
3). (d) Although the number of possible actions (i.e., the
outdegree of node s
i
) is the same as in the previous case,
each action has now a unique response.
game as a directed graph where the game states are nodes
and the possible actions arches (Figure 2). This means that
the greater the outdegree (or fan-out) of a node, the more
freedom the player has. In this graph, the uniqueness of
a response can be measured as the indegree (or fan-in) of
a node. For example, assume that the game plot is divided
into chapters. Typically, the plot lines of the previous chap-
ter are concluded, and many new plot alternatives are intro-
duced. This means that in the graph the beginning state of
the chapter has a large indegree and outdegree. Thus, the
game properties can be analyzed through graph concepts
(e.g., repetitiveness corresponds to cycles in the graph).
In an open game world, the player should have as much
freedom as possible and the responses to the actions should
be appropriate. An obvious way to implement open game
world is simulation. CGs are often likened to simulations,
although there are games that are not simulations: We may
allowthat chess simulates warfare in a highly abstract man-
ner but clearly checkers is not a simulation. Nevertheless,
most games—and especially computer games—are simu-
lations, because a resemblance to real-world objects or ev-
eryday world (even with a slight touch of fantasy) assists
immersion. Indeed, entertainment industry has embraced
military simulations because of the realism they provide
[1].
The main difference between simulations and games is
that games are goal oriented. A flight simulator, for ex-
ample, is not a game in itself but dogfighting with a flight
simulator is. In CGs, a usual approach is to include a story
into the game and, as a consequence, limit the simulation.
This game-as-a-story approach usually contains a linear—
or at most a slightly diverse—plot, where the player has
some freedom only between fixed entry points (i.e., the
game graph converges to a predetermined state from time
to time). Still, most games do not include a story-line nor
impose a sequence of events.
Clearly, the game developers would like to include a
story into the game. The problem is, as we saw earlier,
that the story-generating program must act like a human
dungeon master. It must observe the reactions of the crowd
as well as the situation in the game, and recognize what
pattern fits the current situation: Is the game getting bor-
ing and should there be a surprising plot-twist, or has there
been too much action and the players would like to have a
moment’s peace to rest and regroup?
If we consider pattern recognition, the grand challenge
for the future game development is story generation in an
open game world. Although there have been some efforts
towards this direction, they are more like a set of sepa-
rate missions that have no connection to the dynamic game
world itself. Nevertheless, let us propose a design for such
a CG using medieval world as an example.
Since we aim at telling a story to the human players, we
must ensure that the world around them remains purpose-
ful. We have general plot patterns that we try to recognize
in the history and surroundings of a human player. One of
these pattern could a parental quest: If the player lives with
his or her parents, they can be abducted thus leading the
player on a quest to find them. It does not matter, whether
the player is an heir to the throne or a humble peasant, the
pattern can be found and the story can be set into motion.
We can use lazy evaluation to generate history and non-
player characters (NPCs) as they are needed. If during the
quest the player chances to meet a robber and manages
to befriend with him, we have to line out the history and
the character of the NPC; otherwise, he can simply yell
“Stand and deliver!”, take the player’s money, and vanish
into the night. In the latter case, the story generator could
now recognize the pattern that the player is penniless, on a
quest and wandering aimlessly, and bring into the world a
monk who can tell the player something about the fate of
the missing parents.
Although the story generation focuses on the human
players, there can be large-scale story-lines, which in turn
affect the players. If a war breaks out, the player can sud-
denly find himself or herself in a battlefield. Also, the war
can be incorporated to the player’s story. Perhaps the par-
ents’ capture is due to the hostility between two nations—
or perhaps the reason for it. As we can see from this short
example, the story generator acts like a story-teller: it pro-
vides a sense purpose into the chaotic world.
3.4. Prediction and Production
The phenomena may concern states and events (i.e., time-
dependent relationships). For example, a group formation
can be identified by observing relative locations and ori-
entations of soldiers close together. However, in order to
recognize a forthcoming ambush, one must ponder on the
maneuvers of the troopers.
Let us consider the temporal issues. Utilization of pat-
tern recognition techniques in the case of time series data
is illustrated in Figure 3. Here the world is seen as a gen-
erator that produces events and states, which can be named
with symbols from the alphabet Σ. The current symbol se-
quence S can be used to construct a model for the symbol
stream (i.e., to learn the behavior of the generator). Mod-
eling means recognizing the underlaying dependencies be-
tween the symbols. Typically the dependencies between
the nearby symbols in the sequence are stronger than the
distant ones. Therefore, it is often sufficient to consider
a short term history, called modeling context, of length `
symbols. More accurate model can be achieved by increas-
ing the length ` at the cost of time. There is a number of
ways to model the sequence S, such as Markov chains and
time series neural networks.
In Figure 3, the model gives the probability distribution
for the next symbol in the sequence. This probability dis-
tribution can be used in two different purposes: prediction
and production of forthcoming events and states.
When we are predicting the next symbol in the sequence
S, we select the symbol with the highest probability in the
model at the context S. Utilization of this prediction de-
pends on the game. For example, in a fighting game the
symbols could be movements like kick, hit, or duck. The
generator is the opponent whose actions form the sequence
S. Now we can predict the opponent’s next move and pre-
pare an appropriate counteraction.
The same model structure can be used to produce new
events into the game world. For example, the actions of
the computer controlled actors could be produced by ap-
plying the probability distribution of the model in context
S. By selecting the next action randomly according to the
probabilities of the actions, the behavior of the synthetic
character could seem to be more reasonable. Obviously,
this approach could be utilized in story generation that was
discussed in previous section.
3.5. Other Employments
In addition to AI-related problems, pattern recognition can
be utilized in finetuning the rules of the game world. For
example, in RTS games we could observe that civilizations
A BA A AB B
S =
Prediction
Production
Modeller
a∈Σ: p(a|S)
Selection
by max
_
Generator
A
A BA A AB B
S =
Modeller
a∈Σ: p(a|S)
Selection by
random [0..1]
_
A
Figure 3: Model over a sequence of symbols can be used
in prediction and production. Alphabet Σ = {A, B}.
are technologically unbalanced, or recognize the choke
points both in the technology tree and in the terrain.
A good game has an intuitive interface that is easy to
learn. The interface should adapt dynamically to the needs
of a player. For example, in critical situations—which
should be recognized— the player ought to be able to have
more detailed control.
Likewise, living in a game world should be natural. Tu-
torials are a convenient method for illustrating the interface
and the game world to the player. Hence, it can reduce
the learning curve significantly. To be meaningful, tutori-
als should be adaptive to the skills ofa player. This requires
recognition of modus operandi of the player. This knowl-
edge could be utilized also to keep the game challenging
by proposing suitable difficulty levels for the player.
4. Discussion
Let us review Figure 1 in the light of CGs. The world rep-
resents now the game world, which exists only virtually
in the computer. For this reason, CGs are in a slightly
different position than most of the other pattern recogni-
tion applications. Whereas in other areas the developers
can rely on the experts to help the development process, in
CGs they are the experts themselves; the world is not given
but generated by the very same persons. Also, we get so
much information from the virtual environment that it is
hard to discern which is useful for pattern recognition. For
instance, in a RTS game the chosen strategy can be based
on minutiae from the world but, in turn, that will slow down
the pattern recognition process.
On the other hand, even in the computer games all pat-
terns to be recognized do not originate from the virtual
world itself. To implement challenging computer oppo-
nents we need to recognize the behavioral pattern of a hu-
man player, also. This is especially difficult due to hu-
man’s ability to create extraordinary actions. Indeed, the
lack of challenging and intelligent computer players is one
of the main reasons for the popularity of the multiplayer
games. Further, to provide a good game interface to a hu-
man player, we should observe the actions of the player
and try to recognize the needs of information and interac-
tion (i.e., relevant information is easily available and in-
teraction with the game is smooth and intuitive). In fact,
research results from augmented reality could be utilized
in the development of user interfaces for CGs.
As we saw in Section 3.1, a pattern recognition system
can be composed of several levels of detail. This could be
utilized in design of difficulty levelsfor human players with
varying skills. Because pattern recognition is separated on
several levels, we can easily select how much information
is provided to a higher level recognizer—even leave out
some recognition parts. Similarly, different personalities
of the synthetic players could be realized by filtering the
recognized information.
Pattern recognition methods are widely used in CGs [6].
However, it is not always clear for the game developers
where different pattern recognition methods are applicable.
For example, the seems to be confusion where neural net-
works or fuzzy logic is best suited. Also, pattern recog-
nition should be seen as reusable and modular software
component whose correct design and implementation im-
prove the overall efficiency of the game development pro-
cess. Luckily, the gaming community is turning towards
the academic community [4], and we—as members of the
academia—should welcome this development with open
arms. After all, we all want to have better and more en-
tertaining CGs.
5. Conclusion
We introduced computer games as a quaint application area
for pattern recognition. Pattern recognition is an important
part of whole AI system by providing high-level informa-
tion from the game world to the decision-making system.
Thus, the task of pattern recognition is to extract relevant
information from the world. We discussed how pattern
recognition in CGs can be viewed from several perspec-
tives.
References
[1] M. Capps, P. McDowell, and M. Zyda. A future for
entertainment-defense research collaboration. IEEE
Computer Graphics and Applications, 21(1):37–43, 2001.
[2] G. Costikyan. I have no words & I must design: Toward a
critical vocabulary for games. In F. M
¨
ayr
¨
a (editor),
Computer Games and Digital Cultures Conference
Proceedings, pp. 9–33, Tampere, Finland, June 2002.
[3] Federation Internationale de Football Association. Laws of
the game. Web page, Nov. 2002.
http://www.fifa.com/refs/laws E.html.
[4] International Game Developers Association. Web page, Nov.
2002. http://www.igda.org/.
[5] J.E. Laird and M. van Lent. Human-level AI’s killer
application: interactive computer games. AI Magazine,
22(2):15–25, 2001.
[6] S. Rabin, editor. AI Game Programming Wisdom. Charles
River Media, Hingham, MA, 2002.
[7] RoboCup. Web page, Nov. 2002. http://www.robocup.org/.
[8] R. Schalkoff. Pattern Recognition: Statistical, Structural
and Neural Approaches. John Wiley & Sons, New York, NY,
1992.
... Pengenalan pola digunakan untuk pengambilan keputusan dalam hal prediksi desain dari sebuah permainan dan untuk tahapan pembuatan permainan (produksi). Tugas pengenalan pola adalah untuk mengekstraksi informasi yang relevan, dalam permainan diimplementasikan untuk merekam strategi, taktik, tingkat kesulitan, pergerakan musuh buatan, dan pergerakan player, sehingga dapat membuat/merekomendasikan alur cerita baru dalam permainan sesuai hasil perekaman pola [7]. Selanjutnya menurut [8] pengenalan pola dalam permainan digunakan lebih teknis dalam mengidentifikasi strategi player selama memainkan permainan, hasil pengenalan pola dari strategi player tersebut dimanfaatkan sebagai pengetahuan untuk meningkatkan pola dan kecerdasan gerak-gerik musuh. ...
Article
Full-text available
Dalam bermain sebuah game, faktor motivasi dalam bermain game sangat memiliki peran penting terhadap keberlangsungan dari misi game tersebut untuk diselesaikan. Penelitian ini bertujuan untuk melaporkan hasil pengenalan pola player dan tingkatan motivasi player pada game yang telah dibangun bernama Ghost Traditional Game. Game tersebut menanamkan unsur Artificial Intelligence (AI) dengan jenis Decision Tree dan Finite State Machine berbasis desktop, terdiri dari sepuluh level dengan kategori kemudahan dan kesulitan yang telah diatur sedemikian rupa pada tahapan Game Design. Metode yang digunakan adalah matching player pattern detection yang direkam melalui log data dan motivation questionnaire technique. Sejumlah 30 orang player mengujicoba game ini, dengan tingkat usia yang berbeda-beda antara 12-22 tahun. Hasilnya adalah untuk pengenalan pola player terhadap iterasi masing-masing level yang digunakan pada saat game testing cenderung lebih cocok sama dengan prediksi pola pada game design, untuk motivasi player dalam menyelesaikan misi game dengan kategori sangat setuju, dan rata-rata waktu yang dibutuhkan oleh player untuk menyelesaikan misi game adalah 60 menit.
... As part of AI methods, Kaukoranta et al. (2003) discussed pattern recognition in the context of computer games and its role in extracting relevant information from the game environment, cluster this information into patterns, and pass this information to the decision making system to choose appropriate action. They stated that, in real-time strategic games, recognizing the behavioral pattern of a human player is very important aspect in implementing a challenging computer opponent. ...
Article
Full-text available
A modern computer-based simulation tool (WaterMan) in the form of a game for on-farm water management was developed for application in training events for farmers, students, and irrigators. The WaterMan game utilizes an interactive framework, thereby allowing the user to develop scenarios and test alternatives in a convenient, risk-free environment. It includes a comprehensive soil water and salt balance calculation algorithm. It also employs heuristic capabilities for modeling all of the important aspects of on-farm water management, and to provide quantitative performance evaluations and practical water management advice to the trainees. Random events (both favorable and unfavorable) and different strategic decisions are included in the game for more realism and to provide an appropriate level of challenge according to player performance. Thus, the ability to anticipate the player skill level, and to reply with random events appropriate to the anticipated level, is provided by the heuristic capabilities used in the software. These heuristic features were developed based on a combination of two artificial intelligence approaches: (1) a pattern recognition approach and (2) reinforcement learning based on a Markov decision processes approach, specifically the Q-learning method. These two approaches were combined in a new way to account for the difference in the effect of actions taken by the player and action taken by the system in the game world. The reward function for the Q-learning method was modified to reflect the suggested classification of the WaterMan game as what is referred to as a partially competitive and partially cooperative game.
... We consider our player modelling approach as one primarily aimed at facilitating player-centred design with a method that supports ease of interpretation and adaptation of results. User-centred design (Katz-Haas 1998) arose in domains other than entertainment such as Technology Enhanced Learning (TEL). These systems in fact often use the game-play paradigm of interaction (Cowley et al. 2011; Malone 1980; McGinnis et al. 2008), and rely heavily on user modelling techniques (Beal et al. 2002; Zhou and Conati 2003). ...
Article
Full-text available
The power of using machine learning to improve or investigate the experience of play is only beginning to be realised. For instance, the experience of play is a psychological phenomenon, yet common psychological concepts such as the typology of temperaments have not been widely utilised in game design or research. An effective player typology provides a model by which we can analyse player behaviour. We present a real-time classifier of player type, implemented in the test-bed game Pac-Man. Decision Tree algorithms CART and C5.0 were trained on labels from the DGD player typology (Bateman and Boon, 21st century game design, vol. 1, 2005). The classifier is then built by selecting rules from the Decision Trees using a rule- performance metric, and experimentally validated. We achieve ~70% accuracy in this validation testing. We further analyse the concept descriptions learned by the Decision Trees. The algorithm output is examined with respect to a set of hypotheses on player behaviour. A set of open questions is then posed against the test data obtained from validation testing, to illustrate the further insights possible from extended analysis.
... In computer games, pattern recognition can act in different roles [Kaukoranta03]. These roles depend on ...
Chapter
Full-text available
As game worlds resemble the real world more closely, their increasing complexity requires effective and reliable pattern recognition. Computer games encompass a variety of problems involving pattern recognition including enemy evaluation and prediction, coaching, group coordination, terrain analysis, and learning. The task of pattern recognition is to abstract relevant information from the game world and, based on the retrieved information, construct concepts and deduce patterns for the use of higher level reasoning and decision-making systems. The plan of this article is twofold. Firstly, we analyze what is required from pattern recognition in computer games. We conclude that it can act in different roles, which in turn affect the choice of a method and its implementation. Secondly, we review a branch of pattern recognition techniques arising from soft computing. Our intention is to clarify where these methods should be used.
Article
Full-text available
The current game industry around the world is one of the fastest growing industries. One gaming genre that is very popular is the real-time strategy games. However, current implementations of games apply extensive usage of FSM that makes them highly predictable and provides less replayability. Thus, this paper looks at the possibility of employing case-based plan recognition for NPCs so as to minimize their predictability. The paper also looks into possible representation adaptions to minimize the resource requirement to maintain the possibility of deployment in mobile devices.
Conference Paper
Pattern recognition has been successfully used in different application areas, its application on identifying playerpsilas strategy during the gameplay which is called player strategy pattern recognition (PSPR), is another interesting area. PSPR can greatly improve game AIpsilas adaptability, and as a result the entertainment of game is promoted. In this paper, pac-man game is used as a test-bed. Kernel classifier of radial basis function (RBF) is chosen to analyze off-line data from gamers who are choosing different strategies, in other words the classifiers are trained with sample data from players using different strategies. The method attempts to use the trained classifier to predict strategy pattern of a future player based on the data captured from its gameplay. This paper presents the basic principle of the PSPR by using the RBF theoretic approach and discusses the results of the experiments.
Conference Paper
Full-text available
Abstract Although,one ,of the ,fundamental ,goals ,of AI ,is to understand and develop intelligent systems that have all of the capabilities of humans, there is little active research directly pursuing ,that goal. We propose ,that AI for interactive computer,games is an emerging,application area inwhich,this goal of human-level ,AI can successfully be pursued. Interactive computer ,games ,have ,increasingly complex,and realistic worlds and increasingly complex,and intelligent computer-controlled characters. In this paper, we further motivate our proposal of using interactive computer games, review previous research on AI and games, and present the different game,genres and the roles that human- level AI could play within these genres. We then ,describe the research issues and AI techniques that are relevant to each,of these ,roles. Our conclusion ,is that ,interactive computer,games,provide a rich environment,for incremental research on human-level AI.,
Article
Full-text available
Although one of the fundamental goals of AI is to understand and develop intelligent systems that have all of the capabilities of humans, there is little active research directly pursuing that goal. We propose that AI for interactive computer games is an emerging application area in which this goal of human-level AI can successfully be pursued. Interactive computer games have increasingly complex and realistic worlds and increasingly complex and intelligent computer-controlled characters. In this paper, we further motivate our proposal of using interactive computer games, review previous research on AI and games, and present the different game genres and the roles that humanlevel AI could play within these genres. We then describe the research issues and AI techniques that are relevant to each of these roles. Our conclusion is that interactive computer games provide a rich environment for incremental research on human-level AI.
Article
Full-text available
The Department of Defense and the entertainment industry are combining their expertise in a drive to exploit advances in technology toward the mutual benefit of each group.
AI Game Programming Wisdom
  • S Rabin
S. Rabin, editor. AI Game Programming Wisdom. Charles River Media, Hingham, MA, 2002.
I have no words & I must design: Toward a critical vocabulary for games [3] Federation Internationale de Football Association. Laws of the game
  • G Costikyan
G. Costikyan. I have no words & I must design: Toward a critical vocabulary for games. In F. Mäyrä (editor), Computer Games and Digital Cultures Conference Proceedings, pp. 9–33, Tampere, Finland, June 2002. [3] Federation Internationale de Football Association. Laws of the game. Web page, Nov. 2002. http://www.fifa.com/refs/laws E.html.