Multi-agent planning and simulation

Pommerman

A Java implementation of Pommerman for research with statistical forward-planning agents, developed in collaboration with DSTL.

Contact me about this project

Why this project is interesting

Pommerman provides a competitive, partially observable multi-agent environment for evaluating planning methods. The research needed a Java environment compatible with its agent implementations and configurable scenarios.

The problem

The available Pommerman framework was written in Python, while the statistical forward-planning research stack used Java. The project required a compatible implementation and scenario modifications for controlled experiments.

Approach

Developed a Java version of the Python Pommerman framework and modified scenarios to support integration with statistical forward-planning agents. The project followed an earlier OpDST collaboration.

My contribution

For Pommerman, I worked on the Java framework and scenario modifications. In the earlier OpDST collaboration with Queen Mary, the Alan Turing Institute and DSTL, my contribution covered framework engineering, experimental design, and results analysis and reporting.

Outcomes and evidence

  • A Java Pommerman environment enabled integration with statistical forward-planning agents.
  • The associated study compared MCTS and RHEA in multiple Pommerman settings.

Skills and methods

  • Java
  • Python
  • Planning and search
  • Multi-agent systems
  • Game AI
  • Experimental evaluation
  • Algorithm design
  • Object-oriented design patterns
  • User interface design

Related work and links

Back to portfolio