See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/325368698

Comparative Analysis of Pathfinding Algorithms A *, Dijkstra, and BFS on

Maze Runner Game

Article · May 2018

DOI: 10.30645/ijistech.v1i2.7

CITATIONS READS

10 3,895

4 authors:

Silvester Dian Handy Permana Ketut bayu yogha Bintoro

Trilogi University Trilogi University
31 PUBLICATIONS   21 CITATIONS    29 PUBLICATIONS   15 CITATIONS   

SEE PROFILE SEE PROFILE

Budi Arifitama Ade Syahputra

Trilogi University Trilogi University
13 PUBLICATIONS   16 CITATIONS    11 PUBLICATIONS   12 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Cryptography View project

Optimization View project

All content following this page was uploaded by Budi Arifitama on 29 June 2018.

The user has requested enhancement of the downloaded file.

 International Journal Of Information System & Technology
Vol. 1, No. 2, (2018), pp. 1-8

Comparative Analysis of Pathfinding Algorithms A *, Dijkstra,

and BFS on Maze Runner Game
Silvester Dian Handy Permana1, Ketut Bayu Yogha Bintoro2, Budi Arifitama3, Ade
Syahputra4
1,2,3,4 Informatics
Study Program, Faculty of Creative Industry and Telematics,
Universitas Trilogi
[email protected], [email protected], [email protected],
[email protected],

Abstract
Maze Runner game is a game that requires pathfinding algorithm to get to the
destination with the shortest path. This algorithm is used in an NPC that will move from
start node to destination node. However, the use of incorrect algorithms can affect the
length of the computing process to find the shortest path. The longer the computing
process, the longer the players have to wait. This study compared pathfinding algorithms
A *, Dijkstra, and Breadth First Search (BFS) in the Maze Runner game. Comparison
process of these algorithms was conducted by replacing the algorithm in the game by
measuring the process time, the length of the path, and the numbers of block played in the
existing computing process. The results of this study recommend which algorithm is
suitable to be applied in Maze Runner Game.

Keywords: Pathfinding, A*, Dijkstra, BFS, Maze Runner

1. Introduction
These days, AI have spread through the game world. Some games have started
implementing AI in its game. Varying from action games, adventure, action adventure,
RPG, simulation, strategy, sports, to idle game. AI spreads in all kind of growing game
genre. The use of AI can bring revenue from the game. Players can feel challenged and
buy some of the facilities provided by the game. It is not rare for the game that use AI to
be better than games that does not use AI [1].
Maze Runner game is a labyrinth created by players. The player have to block the path
of the NPC that has been given pathfinding algorithm with Tetris-like block. Tetris itself
has a block like letter I, J, L, O, S, T, and Z in the game [2]. Players are provided with
various blocks with these letters to block the course of the NPC. The more blocks
traversed by the NPC, the more score the player will get. Therefore, the NPC needs the
fastest path to the destination node[3].
Pathfinding algorithm is required to determine the fastest path possible. There are
several algorithms to find the fastest path. The algorithms are A*, Dijkstra, and Breadth
First Search (BFS). These algorithms are the best pathfinding algorithms [4]. Each of
these algorithms has their own weaknesses and strengths in the process of determining the
fastest path. This research will find the best algorithm for the Maze Runner game.
The A Star or A* algorithm is one of the search algorithms that analyze inputs,
evaluates a number of possible paths and generate solutions. The A* algorithm is a
computer algorithm that is used extensively in graph traversal and the search for paths
along with the efficient path planning process around the points called nodes [5].
Dijkstra’s algorithm uses the principle of greedy, in which each step is selected with the
minimum weights that connect a selected node with another unselected node [6]. While
Breadth First Search is a method that performs a wider search that extends a node pre-
orderly, extending a node and then extending all neighbors of the last node. After that,
extend the unextended nodes and is neighboring with the extended nodes, and so on [7].

ISSN : 2580-7250
Copyright ⓒ 2018 IJISTECH
 International Journal Of Information System & Technology
Vol. 1, No. 2, (2018) pp. 1-8

The comparison process in this algorithm was conducted by measuring 3 variables of

pathfinding computation. These three variables are the measurement of the process time,
the length of the path, and the numbers of block played in the existing computing process.
In the process of measuring time, time will be measured in milliseconds (ms). The
calculated time is the time needed from start node to destination node. The time will start
when NPC starts running. The length of the path is measured by counting the number of
blocks traversed by the NPC. Blocks played in the computation process are calculated
from the number of blocks (nodes) counted before the pathfinder determines its path.
The purpose of this research will determine the best algorithm to be applied in this
Maze Runner game. The algorithm which will be recommended to be applied in the game
is considered to be taken from various aspects of the comparison test results. The results
of this research are expected to help game developers to implements the best algorithm
for making maze-based games.

2. Rudimentary
Research conducted by Aqsa Zafar et al with the title, “Analysis of Multiple Shortest
Path Finding Algorithm in Novel Gaming Scenario” discusses the comparison of 5
pathfinding algorithms. The five algorithms are Breadth-First Search (BFS), Depth-First
Search (DFS), Best-First Search, Dijkstra’s Algorithm, and A* Algorithm. Although
using these five pathfinding algorithms, the discussion in the research only compare
Dijkstra's Algorithm and A * Algorithm with comparative table. In the discussion, game
images from Age of Empire and Civilization V are provided but not yet clearly explained
about the use and the application of the game. What can be drawn from this research is
the attention to the A* algorithm make it seen as the best algorithm compared to the
others [8].
Research with the title, “Pathfinding Algorithm Efficiency Analysis in 2D Grid” done
by Zarembo and Sergejs compares A*, BFS, Dijkstra, HPA*, and LPA* in two-
dimensional sized grids. BFS has the slowest results. This result can be explained by the
fact that the algorithm's operating principle is very simple and it does not use any
heuristics. Dijkstra's algorithm is faster than BFS, but slower than the other algorithms.
A* and LPA* have the same value of performances. LPA* is faster in smaller grids (64,
128, 256), but A* is faster in larger grids (512, 1024). Which leads to the conclusion, that
LPA* is better suited for smaller pathfinding problems, while A* is better used to solve
larger problems [9].
Research conducted by Singhal and Harish with the title, “A Review of the Paper of
Navigation and Pathfinding using Mobile Cellular Automata” describes pathfinding,
cellular automata and software agents. This research studied the pathfinding category and
how is it used in finding the distance between 2 nodes. This research focuses on cellular
automata and for complex modeling systems and computing tools. This research uses
Breadth-first search, Dijkstra's algorithm, and A* to find the distance between the cellular
automata [10].
The study titled, “Pathfinding car racing game using dynamic pathfinding algorithm
and algorithm A*” conducted by Sazaki, et.al. used pathfinding algorithm for NPC to race
against player in car racing game. The pathfinding method used by NPCs in this game is
A* algorithm to find the shortest path on the track and combined with Dynamic
Pathfinding Algorithm to avoid static or dynamic obstacles in its path. The experimental
results in this study show that the combination of both methods can be implemented well
in car racing games with the track conditions is blocked with static obstacles. While
moving on the track with dynamic obstacles, the combination of both methods passes
through the course only under certain conditions [11].

2
 International Journal Of Information System & Technology
Vol. 1, No. 2, (2018) pp. 1-8

3. Research Methodology
The problem appointed in this research is the shortest and most efficient route search in
Simulation Game especially Maze Runner. The research methodology used in this
research is a comparative test using 3 levels in a Maze Runner game with the same
obstacle positions for each tested algorithm. The case measured in this comparative test
are the process time, the length of the path, and how many blocks being played in the
existing computational process. In this research, researcher will use 3 different levels with
different Tetris blocks at each level. After completing the test with 3 different levels, it
can be decided which algorithm is best to be implemented in this Maze Runner game.

4. Result and Discussion

Maze Runner is a game that asks its players to block the path of NPC (Non Player
Character) from the start node to destination node. This game has rules and freedom in the
game. Players are prohibited from fully blocking (closing access) the path or imprisoning
the NPC at the start node, this is so that the NPC can run from the start node to the
destination node. Players can only interfere with course taken by the NPC. Players are
given a limited number of obstacles to be arranged freely within the maze. The obstacles
are Tetris block that looks like the letters I, J, L, O, S, T, and Z. The obstacle can be
arranged by dragging the obstacles from its column to the arena / maze. Before being
placed into the maze, players can also rotate the obstacle by 90 degrees with the rotate
button. After all the obstacles are placed, players can press the GO button to give
command for the NPC to walk towards the destination node. Figure 1 shows the Maze
Runner game that asks its players to place all of the obstacles into the given map.

Figure 1. Maze Runner Interface

Maze Runner game has various levels. Each level has different obstacles and different
score requirement to pass the level. The Next button will be enabled if the player can use
all the obstacles and reach the scores required after pressing the GO button. When the GO
button is pressed, the NPC (Green box) will then search for the shortest path to the
destination node (Red box) with the applied algorithm. The score will be calculated from
the number of squares used by the NPC as the path to the destination node.
In this research, we compared the use of the A*, Dijkstra, and Breadth First Search
(BFS) algorithms applied to the NPC to find a way to the destination node. This research
will use level 1, 2, and 3 to do comparison for each algorithm with the same obstacle
layout at each level. Experiments conducted at level 1, 2, and 3 will arbitrarily place all of

3
 International Journal Of Information System & Technology
Vol. 1, No. 2, (2018) pp. 1-8

the obstacles between the start node and the destination node in order to block the path of
the NPC.

Figure 2. Pathfinding of Maze Runner Game at Level 1 using A* (a),

Dijkstra (b), and BFS (c) algorithms

Figure 2 above is the computation results for Maze Runner game at level 1. Obstacles
are arranged in such a way that the path of the NPC, which should only be diagonally
straight to the destination, becomes slightly turned aside. Figure 2a shows that the NPC
moves using A* and 2b algorithms shows that the NPC moves using Dijkstra algorithm.
While Figure 2c shows that the NPC moves using BFS algorithm. The calculation results
of these three algorithms can be seen in table 1 below.

Table 1. Comparison of Algorithms Computation on the First Level of

Maze Runner Game
Components A* Algorithm Dijkstra’s BFS Algorithm
Algorithm
Length 38 38 38
Time 0.3500 ms 0.3000 ms 0.8000 ms
Computed Blocks 323 blocks 738 blocks 738 blocks

From Table 1 above, the same results of 38 blocks traveled were obtained from the
distance traveled by NPC to the destination node. With these three algorithms, players
will get a score of 38. The computational time required by Dijkstra's algorithm is faster
than A* and BFS. A* has 0.0500 ms longer computation time than Dijkstra. The
computational process of A* algorithm is slightly slower than the Dijkstra and BFS
algorithms. It can be deduced that A* algorithm consumes less memory for its
computation process than any other algorithm.

At level 2 of the Maze Runner game, players are given more obstacles to interfere with
the process of finding the shortest path for the NPC, but the distance between both nodes
is closer. Players are given 17 pieces of Tetris-like block. There are 5 pieces of block that
looks like the letter O, 6 pieces looks like the letter I, 4 pieces looks like the letter L, a
single piece looks like the letter J, and another single piece that looks like the letter Z. At
level 2, the minimum score required to unlock level 3 is 30 points. This means that it
takes a minimum of 30 NPC steps to achieve that score. The three algorithms are tested
after all the obstacles are placed in the maze. Pathfinding performed by the three
algorithms can be seen in Figure 3.

4
 International Journal Of Information System & Technology
Vol. 1, No. 2, (2018) pp. 1-8

Figure 3. Pathfinding of Maze Runner Game at Level 2 using A* (a),

Dijkstra (b), and BFS (c) algorithms

In Figure 3 above, A* algorithm has a path similar to the BFS algorithm. Although
similar, the path taken by A* algorithm is closer to the border of the placed obstacles
while the BFS algorithm is closer to the walls by default. This proves that A* algorithm
extends the arranged obstacles and crawls to the goal.

Table 2. Comparison of Algorithms Computation on the Second Level of

Maze Runner Game
Component A* Algorithm Dijkstra BFS Algorithm
Algorithm
Length 34 34 34
Time 0.4000 ms 0.6000 ms 0.8000 ms
Computed Blocks 415 blocks 618 blocks 618 blocks

From Table 2 above, the same results of 34 blocks were obtained from the number of
squares traveled by the NPC to get to the destination node and thus, give a score of 34 to
the player. The computation time difference between the three algorithms is 0.2000 ms.
The A* algorithm has the fastest computation time than the other algorithms. The
computational process needed for A* algorithm is less than Dijkstra and BFS algorithms
with 415 blocks. Both Dijkstra and BFS algorithms extend 618 blocks to search for the
shortest path.

At level 3 of Maze Runner game, players are given even tougher challenges. The start
and destination node at this third level is located adjacently and only separated by a wall.
The player is required to arrange the obstacles so that the NPC moves at least 50 steps.
Players are given 15 pieces of Tetris-like blocks. There are 6 pieces of block that looks
like the letter O, 2 pieces looks like the letter I, 2 pieces looks like the letter L, a single
piece looks like the letter J, 2 pieces that looks like the letter S, a single piece looks like
the letter T, and another single piece looks like the letter Z. The three algorithms are then
being tested after all the obstacles are placed in the maze. The test results of the
algorithms are shown in figure 4.

5
 International Journal Of Information System & Technology
Vol. 1, No. 2, (2018) pp. 1-8

Figure 4. Pathfinding of Maze Runner Game at Level 3 using A* (a), Dijkstra

(b), and BFS (c) algorithms

In Figure 4 above, A* algorithm has the same path as BFS algorithm. Although both
has the same path taken, A* algorithm has fewer computed blocks than BFS algorithm.
While Dijkstra algorithm has a unique path that is goes through the bottom route even
though the number of blocks on both routes are the same. This is because at the time of
computation, Dijkstra's algorithm meets the last node in the middle of the upper route.
Therefore, the extended path is the bottom path and the algorithm reaches the destination
node with the lower path.

Table 3. Comparison of Algorithms Computation on the Third Level of Maze

Runner Game
Component A* Algorithm Dijkstra BFS Algorithm
Algorithm
Length 58 58 58
Time 1.1000 ms 0.9000 ms 1.0000 ms
Computed Blocks 504 blocks 624 blocks 624 blocks

The number of blocks required for the NPC to reach the destination node on the three
algorithms is the same and the player scores 58. The computation time difference between
the three algorithms is 0.1000 ms. Dijkstra's algorithm is faster than any other algorithm.
In terms of computation, the A* algorithm is slightly slower than Dijkstra and BFS
Algorithms with 504 blocks or a difference of 120 computed blocks. Dijkstra and BFS
algorithms extend 624 blocks to search for the shortest path.

5. Conclusion
Based on the results of this research and evaluation conducted, it can be concluded that:
1. A*, Dijkstra, and Breadth First Search can be used to find the shortest in Maze
Runner Game.
2. A* is the best algorithm in pathfinding especially in Maze game / grids. This is
supported by the minimal computing process needed and a relatively short searching
time.
3. The use of the right algorithm can make the game better in terms of computing
process, memory usage, and computing time.

6
 International Journal Of Information System & Technology
Vol. 1, No. 2, (2018) pp. 1-8

6. Suggestion
Some suggestions that can be taken from this research are:
1. To continue this research with a modified A* algorithm to get better computation
results.
2. Make a test for A* to find the shortest path in other cases such as racing game,
strategy, sport, etc.

References
[1] D. Perez-Liebana, J. Liu, A. Khalifa, R. D. Gaina, J. Togelius, and S. M. Lucas, “General
Video Game AI: a Multi-Track Framework for Evaluating Agents, Games and Content
Generation Algorithms,” 2018.
[2] H. M. Vo, “Optimizating Power Consumption Using Multi-Bit Flip-Flop Technique,” pp.
0–3, 2017.
[3] B. Arifitama and A. Syahputra, “Implementing Augmented Reality on a Multiple Floor
Building as a Tool for Sales Product Knowledge,” Int. J. Technol. Bus., vol. 1, no. 1, 2017.
[4] S. M. Kim, M. I. Peña, M. Moll, G. N. Bennett, and L. E. Kavraki, “A review of
parameters and heuristics for guiding metabolic pathfinding,” J. Cheminform., vol. 9, no.
1, pp. 1–13, 2017.
[5] H. Reddy, “PATH FINDING - Dijkstra ’ s and A * Algorithm ’ s,” pp. 1–15, 2013.
[6] P. Singal Dcsa and R. R. S. Chhillar, “Dijkstra Shortest Path Algorithm using Global
Positioning System,” Int. J. Comput. Appl., vol. 101, no. 6, pp. 975–8887, 2014.
[7] A. A. Jamal and W. J. Teahan, “Alpha multipliers breadth-first search technique for
resource discovery in unstructured peer-to-peer networks,” Int. J. Adv. Sci. Eng. Inf.
Technol., vol. 7, no. 4, 2017.
[8] A. Zafar, K. K. Agrawal, W. Cdr, and A. Kumar, “Analysis of Multiple Shortest Path
Finding Algorithm in Novel Gaming Scenario,” Intell. Commun. Control Devices, Adv.
Intell. Syst. Comput., pp. 1267–1274, 2018.
[9] I. Zarembo and S. Kodors, “Pathfinding Algorithm Efficiency Analysis in 2D Grid,”
Environ. Technol. Resour. Proc. Int. Sci. Pract. Conf., vol. 2, p. 46, 2015.
[10] P. Singhal and H. Kundra, “A Review paper of Navigation and Pathfinding using Mobile
Cellular Automata,” vol. 2, no. I, pp. 43–50, 2014.
[11] A. Sazaki, Yoppy, Primanita and M. Syahroyni, “Pathfinding Car Racing Game Using
Dynamic Pathfinding Algorithm and Algorithm A *,” 3rd Int. Conf. Wirel. Telemat. 2017,
pp. 164–169, 2017.

Authors
Silvester Dian Handy Permana was born in Yogyakarta,
Indonesia on November 26, 1990. He graduated from Universitas
Atma Jaya Yogyakarta - Indonesia, with a degree in Informatics
Engineering and later received a Master’s degree in Information
Technology from Universitas Indonesia - Indonesia.
He currently works for Universitas Trilogi as Informatics Lecturer,
and have experience in lecturing for more than 4 years. His
research interests include Game Application, Game Algorithm and
Complexity, and Multimedia.

7
 International Journal Of Information System & Technology
Vol. 1, No. 2, (2018) pp. 1-8

Ketut Bayu Yogha Bintoro was born in Jakarta, Indonesia on

May 14, 1986. He graduated from Universitas Gunadarma -
Indonesia, with a degree in Information System and later received a
Master’s degree in Computer Science from Universitas Gadjah Mada
- Indonesia.
He currently works for Universitas Trilogi as Informatics Lecturer,
and have experience in lecturing for more than 4 years. His research
interests include Artificial Intelligence, Swarm Intelligence, Multi-
Agent System, Intelligence Agent, and Complex Adaptive System.

Budi Arifitama was born in Rome, Italy on May 23, 1986. He

graduated from Universitas Gunadarma - Indonesia, with a degree
in Informatics Engineering and later received a Master’s degree
in Management Information System from Universitas Gunadarma.
He currently works for Trilogi University as Informatics
Lecturer, and have experience in lecturing for more than 4 years.
His research interests include Augmented Reality, Human
Computer Interaction, and Multimedia.

Ade Syahputra was born in Medan, Indonesia on May 08, 1983.

He graduated from Universitas Gunadarma - Indonesia, with a
degree in Informatics Engineering and later received a Master’s
degree in Information and Communications Technology
Management from University of South Australia, Adelaide.
He currently works for Trilogi University as Informatics
Lecturer, and have experience in lecturing for more than 5 years.
His research interests include Augmented Reality, Human
Computer Interaction, and Supply Chain Management.

View publication stats