The Pommerman 2019 Winner Agent

This repository contans the winner entry for the Pommerman competition of 2019.

Contributors

I've built my agent using _ython's C++ implementation: https://github.com/m2q/pomcpp The repository was forked from his, see README_ython.md as well! The agent's actual decision logic and search method was developed by Márton Görög.

Prerequisites

The code compiles under Windows, but it was mostly developed under Linux using:

• Ubuntu 18.04
• gcc 8 with openmp
• make, cmake

Setup

Download

To fully clone the repository use

$ git clone --recurse-submodules https://github.com/gorogm/pommerman_cpp

Compilation

Under Linux CLion is a recommended environment, which will be able to use the CMakeLists.txt included. Alternatively, you can build from command line:

$ cmake .
$ make

Under Windows Visual Studio is recommended.

For unit tests see _ython's README_ython.md

Running

$ python3 playground/frankfurt_vs_gottingen.py

Uncomment "env.render()" to see the UI.

Project Structure

All of the main source code is in src/* and all testing code is in unit_test/*. The source is divided into modules

-BayesianOptimization (out of use)
-docker (helpers to create docker image)
-playground (this is the python section)
-src
 |-agents (the agent logic, mostly by Márton Görög)
 |-bboard (the C++ engine, mostly by _ython)
-unit_test (game engine testing code)

All environment specific functions (forward, board init, board masking etc) reside in bboard. Agents can be declared in the agents header and implemented in the same module.

All test cases will be in the module unit_test. The bboard should be tested thoroughly so it exactly matches the specified behaviour of Pommerman. The compiled test binary can be found in /bin

Existing Search-based Agents

• frankfurt_agent: this was the winner of Pommerman 2019. This file will be preserved as it is.
• gottingen_agent: forked from frankfurt - currently they are same, but gottingen includes debug features. Try to develop this and compete agains frankfurt.

Most of your enchancements could probably be in gottingen_agent.cpp.

Defining New Agents

Adding a new agent requires modification in a lot of files, mostly due to python-c++ interaction, so for a first step I suggest using the ready-made gottingen_agent. To add a new agent:

• In src/agents copy gottingen or frankfurt to a new agent. I used German city names in alphabetic order (there was once Berlin, Cologne, Dortmund, Eisenach 2018) as a tribute to _ython's work, who lives in Germany.
• Copy class in agents.hpp
• Copy python-C-C++ bridge functions in bboard.cpp
• Copy python agent in playground/pommerman/agents
• Probably useful to copy starter scripts in playground

Testing

See README_ython.md

Citing This Work

Please cite the corresponding report from arxiv.