Starting a game instance

CREW's environments are built to support mulit-player games. Our networking model involves a single server and one or more clients per game. The server instance hosts the environment, runs the simulation, and handles agent policy training, which can be executed on a powerful headless GPU server. Human participants can connect via client instances on less powerful machines, which display synchronized game states and collect human input.

1) To start a game instance, first ensure that Nakama server is running on Docker. If Nakama has not been installed, follow the instructions in Setup Nakama.

2) After Nakama is up and running, simply open a server instance of a pre-built unity game under the directory CREW/crew-dojo/Builds/.

3) Then open a client instance for each human participant. Log in and select match according to the instructions in Log in a game.

Adding AI Agents

AI algorithms are written in python. To enable AI agents in the game, we do not need to manually open a server instance. Instead, the server is automatically openend as part of the python script under Algorithms. Details on how to start a game with AI agents is described in Running an algorithm.