Procedural Content Generation for Computer Games

Labs 2 - Mario

Vojtěch Černý

cerny@gamedev.cuni.cz

Prerequisites

• Java 11 or newer
• Make sure your PATH and JAVA_HOME env. variables point to it
• Windows / Mac / Linux
• Python 3 (NEW)
• (Recommended) IntelliJ IDEA
• Community edition suffices
• As a student you can get Ultimate by just entering your student number

About Mario

• First side-scrolling Mario platformer is Super Mario Bros (1985)
• Made on NES

​

Mario as a framework

• Nintendo is (always) very protective about their creations
• Infinite Mario Bros (2008)
• Java clone, procedurally generates levels
• Markus Persson
• Mario AI Framework (2009)
• Sergey Karakovskiy, Julian Togelius
• customized Infinite Mario Bros
• used for AI / PCG competitions and some research

​

Mario as a framework

• Mario AI Framework, 10^th anniversary (2019)
• Available from http://marioai.org

​

• Improvements:
• Contains a forward model
• Has several AIs and level generators implemented
• Storing levels as human-readable files
• Original NES graphics

​

Technical info

• Levels are standardly 150 x 16 tiles
• Robin Baumgarten’s agent uses A* - should be “somewhat” universal AI
• It isn‘t, really
• Also, he‘s quite slow

The challenges

• How do you recognize a good level in Mario?

The Homework

Create a level generator for Mario with some level of indirectness!

​

Possible options (non-exhaustive list):

• Assemble levels from building blocks
• Use template instantiation
• Use grammars
• Predetermine # of gaps, # of enemies, and generate accordingly

The Repository

Start with the mario-levels repo, which contains a Gradle buildsystem to run Mario

Using IntelliJ IDEA as IDE is highly recommended

Two main files:

• RunGame.kt (runnable by ./gradlew singleRun)
• Runs the generator (set in the file) with an A* agent
• GatherMetrics.kt (runnable by ./gradlew generateCSV)
• Performs a number (default 100) runs, evaluates some metrics into a CSV file

Making it work

• Clone the mario-levels repo
• Try running ./gradlew singleRun
• Try running ./gradlew generateCSV
• Run python in src/main/python/make_graphs.csv
• Check if you have graphs generated out/generator.png

The Task

Part 1 - Create a generator Class

​

• If you use Java, create a generator in:
• src/main/java/levelGenerators
• If you use Kotlin, create a generator in:
• src/main/kotlin/levelGenerators

​

• Observe examples in those folders
• Again, create something you would be happy to put in your game
• Put details into the levels (enemies, pipes, etc.)

The Task

Part 2 – Create an evaluation metric of Mario levels

​

Copy one of

• Java -> src/main/java/metrics/JumpMetricJava.java
• Kotlin -> src/main/kotlin/metrics/JumpMetric.kt

​

and modify it to make your own!

​

Criteria: Make something you would be happy to write about!

​

The Task

Part 3 – Plot your generator performance with your metric

​

To plot graphs, all that should be needed is to:

• Add you metric in src/main/kotlin/GatherMetrics.kt
• Generate CSV in src/main/python/data by running

./gradlew generateCSV

• Run plotting in python in src/main/python/make_graphs.py

​

Plot graphs of linearity vs. your metric on your, notch and benWeber generators (see slide 18)

Discuss your results

• Interpret what the graphs are showing
• Which generator seems best?

Criteria

• Create a generator with some level of indirectness
• The levels should be beatable
• You may try to use Baumgarten agent to validate
• (but even he fails on some fairly trivial obstacles)
• Simple agent by Sergey Karakovski should not be able to finish them

​

Points

Level generator – 2 points (indirect representation, good difficulty)

Metric and graphs to evaluate – 1 points

​

1 bonus point – create a level which you can win, but Robin Baumgarten agent does not

​

bonus points are added to exam points, but do not count towards the 13 points required for lab credit

Acceptable solution:

made from building blocks

Mine

Notch

Ben Weber

Acceptable solution:

made from building blocks

Another option – Your experiment!

• If there’s something cool you would like to try in Mario / platformers generation, let’s discuss it and it may become your homework!

​

• It should still be (at least in part) level generation

​

• Examples:
• Grammar system for platformers

​

Submission details

• Send your solutions to cerny@gamedev.cuni.cz
• Prefix the subject with “NAIL123-HW2”
• Submit:
• two files of Kotlin or Java (human-readable, commented)
• one implementing the generator and the metric
• If you want to use more classes, use internal ones
• at least 3 screenshots of generated levels
• plots of metrics (plot linearity vs. your metric for each of your generator, notch and benWeber)
• comment the results in the plots, why they look like they do
• describe all you have tried, what worked or didn’t, the problems you faced
• how much time you spent with the project (will not be part of evaluation)

​

• Deadline is 30.3.
• If you need more time, talk/write to me before the deadline

Questions / Problems

• You can always e-mail me at cerny@gamedev.cuni.cz

​

• One final recommendation:

Allocate more than one session to work on the project. If you get stuck on some issue, it’s good to clear your head for a bit and return to it later.