Pow-Wow: A Dataset and Study on Collaborative Communication in Pommerman

This is a repository for the paper Pow-Wow: A Dataset and Study on Collaborative Communication in Pommerman.
This repository contains a set of tools that we used to collect dialogue and game-play data.

[Project Website]

Steps to launch a data collection enviornment

We assume there are three machines. One is for a central server that runs a game engine (Server), and other two are for clients that participants use to play the game (Client A and Client B). It's also possible to only have 2 machines, where one of them works as the server and a client at the same time.

Pre-requisites:

• Python 3.6.0+
• Docker (used to run the enemy agents)

1. (Server, ClientA, ClientB): Install the modified Pommerman environment

$ git clone -b multi-play-over-network https://github.com/takuma-ynd/pommerman-network.git  ~/pommerman-network
$ cd ~/pommerman-network
$ pip install -U .

2. (Server): Pull the docker image for enemy agents
   $ docker pull takumaynd/hakozakijunctions-notest
3. (Server): Run python message_server.py on the server.
   This script delivers messages between chat consoles and synchronize them with game engine.
4. (Server): Run python facilitator.py --client_ports 9000 9001
5. (Client A): Forward a client port and messaging port (default: 8000) from server to client.
   For example, ssh -L 9000:localhost:9000 -L 8000:localhost:8000 server.com
6. (Client B): In a similar way, an example port forwarding is:
   ssh -L 9001:localhost:9001 -L 8000:localhost:8000 server.com
7. (Client A): Run python client/client.py --port 9000 and open client/htdocs/chat_9000.html with a web browser
   The web page works as a text-based chat console whose information is synchronized with game engine.
8. (Client B): Run python client/client.py --port 9001 and open client/htdocs/chat_9001.html

Once both players input username and click "Play", the game will automatically start.

Brief description of how the system works

Coming soon...