CivRealm is an interactive environment for the open-source strategy game Freeciv-web, based on Freeciv, a Civilization-like game. Within CivRealm, we provide interfaces for two typical types of agents: tensor-based reinforcement learning agents (see Tensor-agent Repo) based on the Gymnasium API, and language-based agents (see LLM-agent Repo) driven by language models.
We also provide a set of tools for training and evaluating agents, as well as a set of baselines for both types of agents. We hope that CivRealm can serve as a testbed for the development and evaluation of agents that can learn and reason in complex environments. Detailed usage of the CivRealm API can be found in the Documentation.
CivRealm is developed based on freeciv-bot, dependent on freeciv-web and FCIV-NET. In the future, CivRealm will be maintained by BIGAI.
CivRealm requires Python ≥ 3.8
and docker. We have tested on Ubuntu 22.04, Mac OS X, and Windows.
To test CivRealm on http://localhost, please follow the docker installation instructions on https://bigai-ai.github.io/civrealm/getting_started/requirements.html.
After starting the Freeciv-web service, you can connect to the Freeciv-web server via the host machine localhost:8080 using a standard browser.
You can install the stable version of CivRealm by:
pip install civrealm
To install the latest version from the source code or contribute to the project, please follow the instructions below:
git clone [email protected]:bigai-ai/civrealm.git && cd civrealm
pip install -e .
Before testing the installation, please make sure that the freeciv-web service is running. You can check the status of the freeciv-web service by running:
docker ps
You should see a docker container named freeciv-web
running.
To test the installation, run the following command after installation. This will start a single player game against the built-in AIs with the default settings.
test_civrealm
!!! success If the installation is successful, the output should be similar to the following:
```bash
Reset with port: 6300
Step: 0, Turn: 1, Reward: 0, Terminated: False, Truncated: False, action: ('unit', 104, 'move NorthEast')
Step: 1, Turn: 1, Reward: 0, Terminated: False, Truncated: False, action: ('unit', 117, 'move North')
Step: 2, Turn: 1, Reward: 0, Terminated: False, Truncated: False, action: ('unit', 118, 'move North')
Step: 3, Turn: 1, Reward: 0, Terminated: False, Truncated: False, action: ('unit', 119, 'move SouthEast')
Step: 4, Turn: 1, Reward: 0, Terminated: False, Truncated: False, action: ('unit', 120, 'move SouthEast')
```
To test with multiple players, run the following command in a terminal to start the game with player myagent
:
test_civrealm --minp=2 --username=myagent --client_port=6001
Then start another terminal and join the game with player myagent1
:
test_civrealm --username=myagent1 --client_port=6001
The following are some common issues that you may encounter when running the code. If you encounter any other issues, please feel free to open an issue.
-
If firefox keeps loading the page, please try to add the following line to
/etc/hosts
:127.0.0.1 maxcdn.bootstrapcdn.com 127.0.0.1 cdn.webglstats.com
-
If you see the following error when running
test_civrealm
, please see this solution. If this does not solve the problem, please checkgeckodriver.log
for more information.selenium.common.exceptions.WebDriverException: Message: Process unexpectedly closed with status 1
One potential solution on Ubuntu 22.04 is:
sudo apt install firefox-geckodriver ln -s /snap/bin/firefox.geckodriver geckodriver
-
If you see the following error when setting
take_screenshot: True
, it is caused by snap version of Firefox. Please try System Firefox installation.Your Firefox profile cannot be loaded. It may be missing or inaccessible.
-
If the screenshot is not centered on the location of your first unit, it is because you are using multiple displays. Please ensure the Firefox browser for screenshot pops up on your primary display.
Our paper is available on Arxiv. If you find our code or databases useful, please consider citing us:
@inproceedings{qi2024civrealm,
title = {CivRealm: A Learning and Reasoning Odyssey in Civilization for Decision-Making Agents},
author = {Siyuan Qi and Shuo Chen and Yexin Li and Xiangyu Kong and Junqi Wang and Bangcheng Yang and Pring Wong and Yifan Zhong and Xiaoyuan Zhang and Zhaowei Zhang and Nian Liu and Wei Wang and Yaodong Yang and Song-Chun Zhu},
booktitle = {International Conference on Learning Representations},
year = {2024},
url = {https://openreview.net/forum?id=UBVNwD3hPN}
}