Exploring Dynamic Difficulty Adjustment in Videogames

Abstract

Videogames are nowadays one of the biggest entertainment industries in the world. Being part of this industry
means competing against lots of other companies and developers,
thus, making fanbases of vital importance. They are a group
of clients that constantly support your company because your
video games are fun. Videogames are most entertaining when the
difficulty level is a good match for the player’s skill, increasing
the player engagement. However, not all players are equally proficient, so some kind of difficulty selection is required. In this paper,
we will present Dynamic Difficulty Adjustment (DDA), a recently
arising research topic, which aims to develop an automated
difficulty selection mechanism that keeps the player engaged and
properly challenged, neither bored nor overwhelmed. We will
present some recent research addressing this issue, as well as an
overview of how to implement it. Satisfactorily solving the DDA
problem directly affects the player’s experience when playing the
game, making it of high interest to any game developer, from
independent ones, to 100 billion dollar businesses, because of the
potential impacts in player retention and monetization.
Keywords—Dynamic Difficulty Adjustment, Videogames, Artificial Intelligence

Full text

Exploring Dynamic Difficulty
Adjustment in Videogames
Gabriel K. Sepulveda, Felipe Besoain, and Nicolas A. Barriga
Abstract—Videogames are nowadays one of the biggest en-
tertainment industries in the world. Being part of this industry
means competing against lots of other companies and developers,
thus, making fanbases of vital importance. They are a group
of clients that constantly support your company because your
video games are fun. Videogames are most entertaining when the
difficulty level is a good match for the player’s skill, increasing
the player engagement. However, not all players are equally profi-
cient, so some kind of difficulty selection is required. In this paper,
we will present Dynamic Difficulty Adjustment (DDA), a recently
arising research topic, which aims to develop an automated
difficulty selection mechanism that keeps the player engaged and
properly challenged, neither bored nor overwhelmed. We will
present some recent research addressing this issue, as well as an
overview of how to implement it. Satisfactorily solving the DDA
problem directly affects the player’s experience when playing the
game, making it of high interest to any game developer, from
independent ones, to 100 billion dollar businesses, because of the
potential impacts in player retention and monetization.
Keywords—Dynamic Difficulty Adjustment, Videogames, Arti-
ficial Intelligence
I. INTRODUCTION
WHEN someone is playing a videogame, his goal is to
have fun. Game developers must infuse the games with
this fun. The fun factor is composed of four axes: fantasy,
curiosity, player control, and challenge. If kept in balance the
player will stay entertained [1]. The challenge axis is the most
difficult to control because to do so, the game difficulty and
player skill must match.
Setting a single difficulty level fit for every player is not
possible. A solution is to allow the player to choose a difficulty
level. This method has some drawbacks, such as having a
limited set of difficulty levels, creating gaps where players
can fall and having difficulty progression that mismatch player
learning curves. Also, by making the player aware of the
change in difficulty the game experience is affected. Over the
last decade, there have been multiple publications related to the
improvement of this issue, using Dynamic Difficulty Adjust-
ment (DDA) [2]. As a result, there are various algorithms the
developer can use to implement DDA, and choosing the most
appropriate one can be a difficult task. The lack of experience
makes choosing and correctly applying the algorithm harder,
leading to a poor implementation causing conflicts in the
game.
G.K. Sepulveda, F. Besoain, and N.A. Barriga are with the Escuela
de Ingenier´
ıa en Desarrollo de Videojuegos y Realidad Virtual, Facul-
tad de Ingenier´
ıa, at Universidad de Talca. Campus Talca, Chile. (e-mail:
gsepulveda17@alumnos.utalca.cl, fbesoain@utalca.cl, nbarriga@utalca.cl).
Corresponding Author: N.A. Barriga. (e-mail: nbarriga@utalca.cl)
In the following section we will briefly describe the Dy-
namic Difficulty Adjustment problem. Section III will intro-
duce the readers to the assessment of player’s skill levels.
Section IV gives an overall explanation of how a DDA
system works, while section V reviews the implementations
of different approaches to DDA. Finally, we close with a
summary and some pointers for future work.
II. BACKGROU ND
When doing an activity that is neither boring nor frustrating,
the person becomes engrossed in said activity, being able to
perform longer and keep focused on the task. This state of
mind is called the flow channel (flow) [3] and is present in all
fields. This concept was later on introduced in the videogames
area by Koster [4].
In figure 1, it’s possible to note that when the difficulty of
the game is higher than the players skills the activity becomes
frustrating pushing the player into a state of anxiety. In contrast
when the player skills are higher than the difficulty, the game
is too easy, pushing the player into a state of boredom. When
neither of those happen, the user is faced by a challenge whose
difficulty level matches the player’s skill, enabling him to enter
the flow. Providing a series of challenges allows the player to
stay in the flow for longer periods of time.
It is important to note that taking breaks between the chal-
lenges will prevent overwhelming the player. By alternating a
series of constant challenges with break times it’s possible to
create a game that enthralls the player and keeps him playing.
Fig. 1. Flow Channel
CHILECON 2019, October 29-31, Valpara´ıso, Chile
978-1-7281-3185-6/19/$31.00 c
2019 IEEE 1

Hunicke proposes the creation of a system with techniques
for stochastic inventory management that periodically exam-
ines the player progress and dynamically adjust the difficulty
of the game challenges to adjust the player’s overall experi-
ence [5]. Later on, he observes that a DDA system can improve
the player experience without the need for any sophisticated
AI [6].
DDA is an AI-based system that allows the change of
attributes and behaviors within the game in runtime. DDA
measure the player performance and change the game dif-
ficulty to match the player skills. As a result it creates the
challenge used to guide the player into the flow zone. A good
DDA must be able to track the player skill level and adapt to
it. Changes in difficulty must follow the player learning curve
and go unnoticed by the player [2].
A DDA system that fulfills these requirements can increase
the player confidence in his chances to beat the game when
presented with a hard challenge [7] and the core engagement
of the player, resulting in an increased gameplay duration [8].
III. ASS ES SI NG PL AYER SKILLS
The first step needed to implement a DDA system is an
evaluation of the player’s performance. Through a set of
predefined variables it’s possible to asses if the difficulty is
fit for the player. This is done by comparing the current in-
game value of this variables with their expected value.
A. Variables
In order to choose the variables used to evaluate the player
performance it is important to have a clear notion of what is
considered as a failure, a success or a skill that the player
needs to develop within the game. Usually, they’ll correspond
to game rules and win or loss conditions [9].
A good example is given in a study using Pac-Man as a test-
bed. By measuring the number of hits on the maze walls, the
number of keys pressed and the number of direction switches,
it is possible to have an idea of what the player skills are [10].
Also the lives lost and pills collected versus time are good
indicators of failure and success rate.
Another example is given by the analysis performed to the
Multiplayer Online Battle Arena (MOBA) game Defense of
the Ancient (DotA)1In this case the level reached versus time,
and towers destroyed versus time, are used to measure success,
and deaths versus time, used for failure [11]. Additionally, we
can count the number of minions killed, heroes killed, items
completed and missed abilities, among others, to check the
player skill level.
As in the previous examples, all games can include variables
that indicate the current state of the player and give informa-
tion about his performance, success and failure ratio, and his
learning process. The variables selected will depend on the
specific game in which the DDA system is implemented.
Note that a high number of variables will result in a
more precise difficulty adjustment but will also consume more
memory, in the other hand a low number of variables will
1https://en.wikipedia.org//wiki/Defense of the Ancients
result in poor adjustment. The amount of variables chosen
will depend on the complexity of the game but for most of
the studies a number of three to five variables is enough.
B. Data Collection
For the DDA system to keep track of the chosen variables,
the game has to overwrite their value. The tracking can be
event-triggered or permanent.
Event-triggered tracking is for variables that change when
fulfilling a condition or performing an action. In this case, the
method in charge of triggering the event can call the DDA
system class and change the variable value.
An example of this are the variables used in the previously
mentioned Pac-Man research [10]. The number of hits on
maze walls, number of keys pressed or number of direction
changes must be actualized just when the corresponding action
happens.
On the other hand, permanent tracking is applied to vari-
ables that are always in game. The game has to call the
DDA system class and update the values of the variables in
the game loop. Setting a minimum time between updates is
recommended to reduce processing.
Examples of variables that need permanent tracking are
player health, player gold, and game progress.
C. Reference Point
With the collected data, it’s possible to assess the player
skills using the performance of a player that matches the game
difficulty as a reference point. The chances of finding a player
that perfectly matches the game difficulty are zero. Thus, we
need to find a method that can provide the reference point.
Setting the reference point based on the beliefs of what
it should be is the worst way to do so. Even with years of
experience, it’s unlikely to guess the correct value.
One option is to use an AI agent capable of playing the
game. The data of the AI play-through will be then used as a
reference point. In order to get reliable information the agent
must play the game many times, a bigger number of iterations
gives more precise information. An agent that plays the game
perfectly isn’t useful as it has to imitate a real player, to that
end an agent that can play at a medium level is required
instead.
However, the best option would be to have real player data.
Using the same system to get data the game can be tested with
real players. Using a survey after the testing session, the data
of players that had fun while playing can be used as a sample.
D. Data Analysis
Having set the varables to analyze, the reference point for
each, and being able to save the data from the player play-
through, the system has the essentials needed for the evaluation
of player performance. These evaluations should be made
through the use of methods that return a success, failure or
skill ratio. Just subtracting the number of player deaths from
the expected player deaths returns a non-representative number
in most cases. The evaluation must also happen constantly in
CHILECON 2019, October 29-31, Valpara´ıso, Chile
2

the game, for the difficulty to change at the right times. To
this end, the evaluation of the player performance must happen
once each X number of ticks. A low X value makes the game
more adaptable, but increase the process cost of the game and
might not be needed, selecting the right X value will depend
on the game genre.
Evaluation of player performance can be done by comparing
the player’s current stats with the ideal or reference ones versus
a delta time [10] (equation 1).
Dif f iculty = (N−Z)/D;(1)
Ease = 1 −Dif f iculty;(2)
Let’s take, for example, a situation where the player per-
forms an action Ntimes in a Dperiod of time or ticks, with
the ideal or reference value being Z. Using a value set that
abides by 0≤(N−Z)≤Dwill return a normalized value,
allowing an easier interpretation of the results.
As the enjoyment of a game is greater when the game
is equally hard and easy, the point where these two values
merge is the desired game state. A return value close to 0.5
fulfills this condition. The chances of getting the desired value
are low, that’s why leaving a proper margin is recommended.
The margin must be defined by the game developer based
on the situation. Leaving a 0.1 error margin would leave us
with a range between [0.4, 0.6], where the game difficulty is
acceptable.
This method also allows calculating the player global
proficiency by calculating the average of all variables or a
weighting of them. Consider that it is convenient for the global
proficiency, or player global performance, to be a normalized
value. To this end, the sum of all weights must also be
normalized.
Another method is to assess the player current performance
(Cp) versus the expected player performance at that time
(Ep[t]), which would be our reference point.
P erf ormance =Cp/Ep[t]; (3)
In this manner, a value next to 1 means that the challenge
is appropriate for the player. Same as in the previous method
a error margin defined by the game developer is needed as
getting the exact value is unlikely. For example, setting the
error margin of 0.2 give us a range [0.8, 1.2]. This mean that
values lower than 0.8 indicates the game is too hard and values
greater than 1.2 means the game is too easy.
This method also allows to create a ranking of the player
global performance by adding the results. Both methods can
be modified to better suit the game developer preferences and
needs.
IV. IMPLEMENTATION
There are several different DDA methods proposed in the
literature. We will focus on the more straightforward ones,
with the purpose of giving the reader some insight on how
to implement the DDA system into his projects. For more
information, a recent review of DDA research from 2009 to
TABLE I
DDA APPR OACH ES
Author(s) Year Approach
Hunicke and Chapman 2004 Hamlet System [5]
Spronck et al. 2006 Dynamic Scripting [12]
Pedersen, Togelius, 2009 Single and
and Yannakakis multi-layered perceptrons [13]
Hagelback and Johansson 2009 Reinforcement Learning [14]
Upper Confidence Bound
Li et al. 2010 for Trees and
Artificial Neural Networks [15]
Ebrahimi and Akbarzadeh-T 2014 Self-organizing System and
Artificial Neural Networks [10]
Sutoyo et al. 2015 Metrics [16]
Xue et al. 2017 Probabilistic Methods [8]
Stein et al. 2018 EEG-triggered dynamic
difficulty adjustment [17]
2018 is available [2]. Table I summarizes the approaches found
in the literature.
At the start of the game, there is no data of the player
performance, and so, the developer must set the difficulty
based on the average testing results. Then, the game must
quickly adapt to the player performance. Playable tutorials
are a good opportunity to get early information on the player
performance without having to make him go through any real
challenge. Afterward, the change of difficulty should happen
less often and be smaller, allowing for the growth of the player
skills.
The biggest challenge while changing the difficulty of
the game in real time is avoiding the player noticing the
change. To avoid this, performing the difficulty change at
times where the player is not aware is a must. Such cases
are times when the player is dead, change of scenes, features
or elements the player has not yet seen. If the change is subtle
enough, elements that the player is not currently seeing, non-
perceptible changes as most of the element specific variable
values changes, or minor behavior changes can be performed
without fearing the player to notice it, as long as the changes
don’t happen too often, and aren’t too extreme. There must be
an upper and lower limit for how much a variable can change,
as well as a time threshold for the changes to happen and a
maximum number of changes for each update and stage. The
changes to be performed can be added to a queue as functors
or callbacks to be executed at a given time. Each change can
have a tag that identifies the change. When a change is going
to enter the queue any changes performed with the same tag
will be removed, preventing two changes to affect the same
element, as this is unneeded and can cause problems.
Big changes, such as the size of a room, conspicuous change
of behavior with no reason in the gameplay, and others, must
be executed in load screens, while changing scenes, or in
sections not seen by the player.
These can also be added to the functors or callbacks queue
and be executed when needed. On the other hand, changes to
be performed on zones unseen to the player, but on the same
CHILECON 2019, October 29-31, Valpara´ıso, Chile
3

scene, can be performed immediately to prevent the player for
getting to the zone with the changes undone.
The creation of mathematical functions that tells us how
much the variables should be changed is probably the best
way to perform these changes. The functor can receive as
parameter the proficiency of the player in the corresponding
field and calculate how much each variable must change to
fit the player. The definition of the mathematical functions
must be done with the data collected from the multiple testing
phases. Creating a game with preset difficulties and make a
study where players test all the difficulty levels is advisable.
Each one of the variables selected as an indicator of the
player performance is directly affected by at least one factor
in the game. For example, the death ratio of the player depends
on the damage dealt by the enemies, the chances to evade an
attack, and the aggressiveness of the enemies, among others.
Linking each evaluation variable to a factor is crucial, as this
factors are what is going to be changed in order to adapt the
game difficulty to the player. These factors can be categorized
in three sections, attributes, behaviors and events.
A. Attributes
Changing the value of attributes in the game is the first
and most intuitive of the modifications to be done, and also,
the easiest of them. Even so, the number of attributes that
can create the difficulty trait in one of these variables can
be immense. Specific information, like the given by reports
of event-triggered parameters, can allow discerning how to
properly balance the game. By comparing the player stats with
the enemy who killed him, and the time of the engagement, it
is possible to know if the player died because of the difference
in damage, speed, range, attack ratio or others. There are also
situations where the player died because his health was low
before the engagement, which means that it’s not the current
enemy the reason for his death, but a previous one or the sum
of them. The tracking of spikes in permanent game values
allows a better adjustment. If there was a spike drop in health,
knowing whether it was caused by a single enemy or by a
group attack, can trigger the adjustment of the single enemy’s
stats, or the reduction in the number of enemies. If there was
no spike then maybe the speed at which the waves of enemies
reach the player is too fast and slowing down is required. As
shown in the previous example, the traits of the characters
in the game are not the only thing that can be changed, but
also the number of enemies on the same zone, or even subtle
things, such as the auto-aim range, the time limit for a quick
time event, the dimension of a room, or the pace of the game.
B. Behaviors
Traits in a game can be presented not only as attributes
but also as behaviors. Behaviors are the main component of
complexity in a game and are not exclusive to NPCs (Non-
Playable Characters), in three-in-line games the pieces have
the behavior to self destruct if there are another two equal
pieces at a two tiles distance (vertical or horizontal), in the
same axis (x,y). Modern three-in-line have added new blocks
and behaviors. These behaviors cannot be changed, as they
are the foundation of the game, but not all games have these
restrictions.
In a stealth game, the watch tower can move the light fol-
lowing different patterns, alternating patterns, or just randomly.
The more predictable the pattern is, the easier for the player.
The ability to communicate with other watch towers and to
detect footprints or noises are other behaviors that can be
enabled, disabled, or modified to change the difficulty of the
game. In a MOBA, the tower has the behavior to attack the
enemies, by changing the target priority, the game changes the
difficulty drastically.
C. Events
Events provide an alternative that doesn’t required too many
modifications to the existing game codebase, as does the
change in attributes and behaviors, but their implementation
requires more design work than the previous two, and are
easier to be spotted by the player if the implementation is
poor. Events are predefined occurrences that arise under certain
circumstances. For example, if the player is low on health,
and his performance is low, the next enemy will always drop
a health potion. Conversely, if the player’s health is full, and
his proficiency level is high, the next enemy hit will always
deal critical damage.
V. M ODELS
In this section, we will introduce a small selection of
existing approaches for DDA.
A. Metrics
As mentioned before, it’s important to identify the factors
that directly influence the player performance. In the simpler
use of metrics, a multiplier is applied to the variables that
controls said factors. The initial value for the multipliers is to
be defined by the game developer, it’s recommended to use a
neutral multiplicative at the start of the game as there shouldn’t
be any change on the factors yet. For each relationship between
the variables used to evaluate player performance (EV ) and
the attributes that affect the player performance (F V ), a weight
is defined. Note that the maximum number of weights needed
is of EV ×F V , in case that each attribute affects every
variable, which means adding variables or attributes would
greatly increment the number of weights to define. If possible,
each variables should be affected by a few attributes, and not
all of them in order to reduce the number of weights.
These weights are added to the multiplier in function of
the difference between player performance (P P ) and game
difficulty (GD). The evaluation can be performed through the
use of thresholds [16] or multiplying the weight with said
difference P P
GD .
B. Probabilistic Methods
Probabilistic methods focus on predicting events on the
game through the use of probabilistic calculations and using
the probabilities in a challenge function [8].
CHILECON 2019, October 29-31, Valpara´ıso, Chile
4

The probabilistic calculations are used to get the expected
value of factors that directly affects the player performance
and act accordingly before the player faces the challenge.
As an example, in case the player is reaching a zone with
40% of his health, the probabilistic calculation will get the
expected value of the total damage the player is going to suffer.
This value will be returned to the challenge function that is
going to evaluate whether the challenge is too difficult or too
easy for the player and act accordingly before the player enters
the zone.
Experiments show that the probabilistic method is effective
in games organized in stages [18] or levels [8], as it allows
the AI to know which calculations it has to make based on
the player current location and direction or to precalculate the
values of each stage.
C. Dynamic Scripting
Dynamic Scripting is an online machine learning technique
focused in the modification of behaviors of agents in the game.
Dynamic Scripts (DS) are built from a set of rulebases, one
for each type of agent to be modified. Each time an agent is
created, the associated rulebase is used to create a new script
that controls its behavior. Each rule of the rulebase has an
associated weight that determine the chances of selecting the
rule. The weights are adjusted based on a fitness function that
evaluates the system’s performance [12].
Image 2 shows an example of a Dynamic Scripting system
in action. The character has an action rule-base. Variations
on the same rules are performed to give the algorithm more
options. The DS selects rules from the rule base to create the
behavior script.
In order to have a good performance, a Dynamic Scripting
implementation has to meet certain requirements [12], [19]:
Speed: Algorithms in online machine learning must be com-
putationally fast since they take place during gameplay.
Effectiveness: The created scripts should be at least as chal-
lenging as manually designed ones.
Robustness: The learning mechanism must be able to cope
with a significant amount of randomness inherent in most
commercial gaming mechanisms.
Efficiency: The learning process should rely on a small num-
ber of trials, since a player experiences a limited number
of encounters with similar groups of opponents.
Clarity: Adaptive game AI must produce easily interpretable
results, because game developers distrust learning tech-
niques of which the results are hard to understand.
Variety: Adaptive game AI must produce a variety of dif-
ferent behaviors, because agents that exhibit predictable
behavior are less entertaining than agents that exhibit
unpredictable behavior.
Consistency: The average number of learning opportunities
needed for adaptive game AI to produce successful results
should have a high consistency to ensure that their
achievement is independent both from the behavior of
the human player, and from random fluctuations in the
learning process.
Fig. 2. Dynamic Scripting [20]
Scalability: Adaptive game AI must be able to scale the
difficulty level of its results to the skill level of the human
player.
As Dynamic Scripting is a continually learning AI, it might
keep improving until it reaches a point where it can always
defeat the player. In DDA the objective isn’t to beat the player
but to provide a fitting challenge, so in order to avoid the
AI surpassing the player skills some modifications have been
added to traditional Dynamic Scripting.
Weight Clipping is a technique where a maximum Wvalue
is determined, preventing the weights to grow over W[21].
Normally, a set of rules with a large win rate will result in an
AI with high performance, thus increasing the weights of those
rules and increasing the chances of being selected, resulting
on an AI that keeps defeating the player. By setting a low W
value, the chances of the rule set beaing selected are reduced,
allowing the AI to pick tactics that make him lose against the
player.
Top Culling technique, as Weight Clipping, sets a maximum
Wvalue and allows the weights to grow limitless but weights
with values that surpasses Wwill not be selected to generate
scripts [21].
Another technique used to increase the player enjoyment
is the Adrenaline Rush. This technique is based on the
assumption that the player’s learning rate is high at the start of
the game but drops through it. A learning limit is set to restrict
the weights adjustment, and a maximum player fitness delta
Pis defined. By constantly measuring the player fitness and
keeping track of the previous value a player fitness difference
between the previous and current state can be calculated. When
this player fitness delta drops below Pthe learning limit
of the weights is decreased so the AI doesn’t overgrow the
player [22].
CHILECON 2019, October 29-31, Valpara´ıso, Chile
5

Finally, it’s possible for the Dynamic Scripting’s fitness
function to minimize the difference between it’s performance
and the player’s, rather than to maximize absolute perfor-
mance. This way the AI will adjust the weights to select
rules that increase the chances of the game difficulty fitting
the player skills, instead of the rules that make it more likely
to win.
VI. CONCLUSION
Dynamic Difficulty Adjustment (DDA) is a technique that
allows the developer to give the player a game that adapts
itself to fit him, thus increasing player engagement. In recent
years the amount of research in DDA has increased. As a
result, there are many approaches to implement it, with the
main difference being the method used to change the attributes
and behaviors of the game. Because of that, it is possible to
have a general structure of how to implement the DDA. The
proposed structure is separated into two main parts, measuring
the player proficiency and adjusting the game accordingly,
using the method preferred by the developer.
Implementing a Dynamic Difficulty Adjustment system is
a powerful tool that allows for a better game experience. Its
implementation is based mainly in data collection from test
subjects and the player playing the game, and use of metric and
mathematical functions to define the changes to be performed.
By adjusting variables and behaviors within the game create
challenges fit to the player, but careful planning is needed.
Poor implementation can result in having the opposite effect
and overuse can cause high processing consumption from the
game.
A. Future Work
DDA is still a new field of research and much can be done
to increase its effect in player engagement. New methods for
implementation are always needed but research on optimiza-
tion of already existing ones or creation of tools that enable
an easier DDA implementation are also interesting.
Research can be done to reduce the process required for the
evaluation of the fitness function, in order to be able to add
more variables and create even more precise DDA systems.
As well as the creation of a method that enable a low cost
adjustment of weights allowing the developer to manipulate
more factors of the game.
Integration of Behaviors Trees and Finite-State Machines
to the Dynamic Scripting approach, where all possible three
nodes or states and transitions are preprogrammed, and the
used ones are selected by the DDA system is another promis-
ing research area.
Finally the creation of tools that allows easy DDA imple-
mentation for all game styles and that can be used in the most
popular frameworks or engines for game development, such
as Unity or Unreal, is a field that is still untouched.
REFERENCES
[1] T. W. Malone, “Toward a theory of intrinsically motivating instruction,”
Cognitive Science, vol. 5, no. 4, pp. 333–369, 1981.
[2] Z. Mohammad, “Dynamic difficulty adjustment (DDA) in computer
games: A review,,” Advances in Human-Computer Interaction, vol. 2018,
no. 12, pp. 333–369, 2018.
[3] M. Csikszentmihalyi, Flow: The Psychology of Optimal Experience.
New York : Harper and Row, 2009.
[4] R. Koster, Theory of fun for game design. ” O’Reilly Media, Inc.”,
2013.
[5] R. Hunicke and V. Chapman, “AI for dynamic difficulty adjustment in
games, challenges in game artificial intelligence aaaiworkshop,” 2004.
[6] R. Hunicke, “The case for dynamic difficulty adjustment in games,”
in Proceedings of the 2005 ACM SIGCHI International Conference on
Advances in computer entertainment technology. ACM, 2005, pp. 429–
433.
[7] T. Constant and G. Levieux, “Dynamic difficulty adjustment impact
on players’ confidence,” in Proceedings of the 2019 CHI Conference
on Human Factors in Computing Systems, ser. CHI ’19. New
York, NY, USA: ACM, 2019, pp. 463:1–463:12. [Online]. Available:
http://doi.acm.org.utalca.idm.oclc.org/10.1145/3290605.3300693
[8] S. Xue, M. Wu, J. Kolen, N. Aghdaie, and K. A. Zaman, “Dynamic
difficulty adjustment for maximized engagement in digital games,” in
Proceedings of the 26th International Conference on World Wide Web
Companion. International World Wide Web Conferences Steering
Committee, 2017, pp. 465–471.
[9] M. Hendrix, T. Bellamy-Wood, S. McKay, V. Bloom, and I. Dunwell,
“Implementing adaptive game difficulty balancing in serious games,”
IEEE Transactions on Games, pp. 1–1, 2018.
[10] A. Ebrahimi and M. Akbarzadeh-T, “Dynamic difficulty adjustment in
games by using an interactive self-organizing architecture,” in 2014
Iranian Conference on Intelligent Systems (ICIS), Feb 2014, pp. 1–6.
[11] M. P. Silva, V. d. N. Silva, and L. Chaimowicz, “Dynamic difficulty
adjustment through an adaptive AI,” in 2015 14th Brazilian Symposium
on Computer Games and Digital Entertainment (SBGames), Nov 2015,
pp. 173–182.
[12] P. Spronck, M. Ponsen, I. Sprinkhuizen-Kuyper, and E. Postma, “Adap-
tive game AI with dynamic scripting,” Machine Learning, vol. 63, no. 3,
pp. 217–248, 2006.
[13] C. Pedersen, J. Togelius, and G. N. Yannakakis, “Modeling player expe-
rience in super mario bros,” in 2009 IEEE Symposium on Computational
Intelligence and Games, Sep. 2009, pp. 132–139.
[14] J. Hagelback and S. J. Johansson, “Measuring player experience on
runtime dynamic difficulty scaling in an rts game,” in 2009 IEEE
Symposium on Computational Intelligence and Games, Sep. 2009, pp.
46–52.
[15] Xinyu Li, Suoju He, Yue Dong, Qing Liu, Xiao Liu, Yiwen Fu, Zhiyuan
Shi, and Wan Huang, “To create DDA by the approach of ANN from
UCT-created data,” in 2010 International Conference on Computer
Application and System Modeling (ICCASM 2010), vol. 8, Oct 2010,
pp. V8–475–V8–478.
[16] R. Sutoyo, D. Winata, K. Oliviani, and D. M. Supriyadi, “Dynamic
difficulty adjustment in tower defence,” Procedia Computer Science,
vol. 59, pp. 435–444, 2015.
[17] A. Stein, Y. Yotam, R. Puzis, G. Shani, and M. Taieb-Maimon, “EEG-
triggered dynamic difficulty adjustment for multiplayer games,” Enter-
tainment computing, vol. 25, pp. 14–25, 2018.
[18] C. Segundo, K. Emerson, A. Calixto, and R. Gusmao, “Dynamic
difficulty adjustment through parameter manipulation for Space Shooter
game,” in Proceedings of SB Games, 2016.
[19] P. Spronck, I. Sprinkhuizen-Kuyper, and E. Postma, “Online adaptation
of game opponent ai in simulation and in practice,” in Proceedings of
the 4th International Conference on Intelligent Games and Simulation,
2003, pp. 93–100.
[20] A. Toubman, J. J. Roessingh, P. Spronck, A. Plaat, and J. van den Herik,
“Improving air-to-air combat behavior through transparent machine
learning,” in Proceedings of the I/ITSEC 2014 Conference, 2014.
[21] P. Spronck, I. Sprinkhuizen-Kuyper, and E. Postma, “Enhancing the
performance of dynamic scripting in computer games,” in International
Conference on Entertainment Computing. Springer, 2004, pp. 296–307.
[22] J. J. P. Arulraj, “Adaptive agent generation using machine learning for
dynamic difficulty adjustment,” in 2010 International Conference on
Computer and Communication Technology (ICCCT), Sep. 2010, pp.
746–751.
CHILECON 2019, October 29-31, Valpara´ıso, Chile
6