NUI Galway – UL Alliance First Annual ENGINEERING AND - ARAN ...
NUI Galway – UL Alliance First Annual ENGINEERING AND - ARAN ...
NUI Galway – UL Alliance First Annual ENGINEERING AND - ARAN ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
Generation and Analysis of Graph Structures with an Application to Generate<br />
Levels (Turn Based Patroller-Intruder Games)<br />
David Power and Colm O’Riordan<br />
CIRG, <strong>NUI</strong>, <strong>Galway</strong><br />
d.power1@nuigalway.ie, colm.oriordan@nuigalway.ie<br />
Abstract<br />
This project focuses on automatically testing game<br />
content and consequently designing a tool which<br />
automatically generates game content. This research is<br />
relevant to current game design techniques as it<br />
reduces game testing times and allows for quick<br />
generation of content.<br />
1. Introduction<br />
Current game design fundamentals include<br />
conceptualising a game, planning how this shall be<br />
accomplished, executing this plan, testing the results<br />
and then refining the game [1]. This can be a long<br />
process along with extensive testing needed to be<br />
performed before a game can be completed. One way<br />
for the testing time to be reduced would be to develop<br />
automated testing of the game that could occur during<br />
all levels of the design stage.<br />
Another way to decrease the length of game design<br />
would be to automate level design. This is especially<br />
useful for genres like patrol games, FPS (<strong>First</strong> Person<br />
Shooter), RTS (Real Time Strategy), where levels are<br />
often quite similar with just small changes in terrain<br />
allowing for different gaming experiences. If the<br />
properties of a level were analysed, then levels could be<br />
generated automatically allowing for much more game<br />
content, increasing longevity and enjoyability for users.<br />
In this research, the domain of patrol games is used<br />
develop ways of testing levels automatically and<br />
generating levels automatically based on the results. A<br />
patrol game involves an intruder agent and one or more<br />
patroller agents [2]. The intruder is inserted into a<br />
predefined area in a level and then must make its way<br />
towards another predefined area in the level known as<br />
the goal. The patrollers are constantly patrolling the<br />
level on predetermined paths trying to find the intruder.<br />
An intruder success is when it reaches the goal area<br />
while a failure occurs when a patroller and an intruder<br />
cross paths.<br />
2. Simulator Model<br />
The graphs being tested by the simulator correspond<br />
to levels of a 2D turn based, node based, patrollerintruder<br />
game. A level is represented as an adjacency<br />
matrix and all possible pathways through the graph are<br />
calculated by brute force. The simulator then randomly<br />
generates paths for both the intruder and patroller. The<br />
simulator plays the two agents against each other and<br />
records either a success or failure depending on the<br />
outcome. If the result is a failure, the positions of the<br />
3<br />
agents at failure are also recorded. The agents are<br />
competed against each other for a large sample of<br />
randomly generated paths. The ratio of successes to<br />
failures will show the level of difficulty of the graph.<br />
3. Current Work<br />
By recording the frequency of failure localised to<br />
each node with respect to the total number of failures<br />
occurring throughout the simulation we hope to identify<br />
certain graph characteristics. The levels of difficulty<br />
each of these characteristics bring to a graph need to<br />
documented and assigned and influence value.<br />
The main characteristics identified so far that<br />
influence graphs are connectivity, choke points, chains,<br />
dead ends and small cycles. Choke points for instance<br />
severely restrict pathways through an area of the maze<br />
making safe navigation for the intruder more difficult.<br />
4. Future Work<br />
Using the values described above, we hope to be<br />
able to generate new graphs within a certain level of<br />
difficulty. This will be accomplished by evaluating<br />
different sections of graph and using genetic algorithms<br />
to form a totally new graph from constituent parts.<br />
Furthermore we hope to run these newly generated<br />
graphs through another simulator (3D real time game)<br />
created by a colleague [3]. By comparing the results of<br />
the two simulators we will see if the same<br />
characteristics and values identified in the 2D turn<br />
based environment, hold true for the 3D real time<br />
environment.<br />
5. References<br />
[1] Chris Crawford, The Art of Computer Game Design,<br />
Osborne/McGraw-Hill, Berkeley, CA, 1984<br />
[2] Amigoni, F.; Basilico, N.; Gatti, N.; Saporiti, A.; Troiani,<br />
S.; "Moving game theoretical patrolling strategies from theory<br />
to practice: An USARSim simulation," Robotics and<br />
Automation (ICRA), 2010 IEEE International Conference on ,<br />
vol., no., pp.426-431, 3-7 May 2010<br />
[3] Costello, F. and O’Riordan, C. “An Approach to Providing<br />
Feedback at the Design Phase in Game Authoring Tools”.<br />
GAME-ON 2009: 10 th International Conference on Intelligent<br />
Games and Simulations (ed. Linda Breitlauch). Mediadesign<br />
Hochschule Dusseldorf and EUROSIS, pp 20-23.