Robocode ML

Machine learning for robocode

Project Structure

$PROJECT
|-analysis/     # analytics data
|-battles/       # battle configurations
|-data/          # generated/reprocessed data from robocode
|-doc/           # project documentation
|-installers/    # 3rd-party installers
|-java/          # java source files
|-python/        # python source files (capture/training/analytics/ML)
|-scripts/       # shell scripts
|-README.md

Setup

Note: A copy of robocode is supplied with this repository

$PROJECT is the root of the project folder

1. Install robocode 1.9.3.2:

   • Open a terminal at $PROJECT/scripts

   • Install robocode: run ./setup (This installs robocode within the project folder, if you install like this, you won't need to fix up any maven paths in pom.xml)

2.a Install java dependencies

- https://maven.apache.org/install.html

- JDK 8

2.b Install python dependencies - python3 - python3-venv - python3-pip - python3-tk

3.a Build from command line: ./build

3.b Build from IDE

• IntelliJ (assuming the project root is ./java/)

• Open the ./java roject and make sure you import the pom.xml (you may get a warning about this)

  • Go to Run > Edit Configurations... > Click + (Add New Configuration) > Select Application

  • Configure with the below parameters:

    • Name: random-vs-random.battle (recommend one configuration per)

    • Main class: robocode.Robocode

    • IF you want to configure robocode to automatically find the project robots

      • VM options: -Xmx512M -Dsun.io.useCanonCaches=false -Ddebug=true -DNOSECURITY=true -DROBOTPATH=../java/target/classes

    • ELSE if you want to set the robots path from within robocode (see '4')

      • VM options: --Xmx512M -Dsun.io.useCanonCaches=false -Ddebug=true

    • Program arguments: -battle ../battles/random-vs-random.battle

    • Working directory: /path/to/robocode (from previous step) e.g. $HOME/Projects/RobocodeML/robocode-1.9.3.2

    • Use classpath of module: RobocodeML

  • You will then need to add the Robocode jars to your classpath, (how you do this depends on the version of IntelliJ) In IntelliJ, go to the Project Settings/Project/Modules proprties page Select the dependencies tab Add (click +) JARs or directories Select the libs folder in your Robocode installation - e.g. $HOME/Projects/RobocodeML/robocode-1.9.3.2/libs

  • Make sure you name this configuration (e.g.random-vs-random.battle) and click Apply or OK, otherwise you'll have to start again

  • Run or Debug

• Eclipse: use these instructions

4. (Optional) After setting up the IDE, robocode needs to know where the project robots are.

   • Run robocode cd $PROJECT/robocode-1.9.3.2 then ./robocode.sh.

   • Go to Preferences / Development Options, add path $PROJECT/robocode/target/classes

5. Run from command line

   • Start the capture: (data is saved to ./data/)

     ./capture

   • Run the Training bot vs the Random bot in Robocode:

     ./battle-test-random (headless) or ./battle-test-random-gui

Notes

java.io.FileNotFoundException: ... /org/json/JSONObject.class (No such file or directory)

Ignore this: I suspect robocode's class loader logs this erroneously, but the json module works fine

Resources

• http://robocode.sourceforge.net/docs/robocode/
• http://robowiki.net/wiki/Talk:Wall_Avoidance
• http://robowiki.net/wiki/Wall_Smoothing
• http://robowiki.net/wiki/Wall_Smoothing/Implementations
• http://www.ibm.com/developerworks/library/j-robotips/index.html
  • Anti-gravity movement
  • Predictive targeting
  • Tracking your opponents' movement
  • Circular targeting
  • Dodge bullets
  • Tracking bullets
  • Radar sweeps
  • Polymorphic enemy cache
  • Robocode strategies