Counting 3,834 Big Data & Machine Learning Frameworks, Toolsets, and Examples...
Suggestion? Feedback? Tweet @stkim1

Last Commit
May. 23, 2019
Jan. 8, 2019

- This is a BETA release.

SMAC - StarCraft Multi-Agent Challenge

SMAC is WhiRL's environment for research in the field of collaborative multi-agent reinforcement learning (MARL) based on Blizzard's StarCraft II RTS game. SMAC makes use of Blizzard's StarCraft II Machine Learning API and DeepMind's PySC2 to provide a convenient interface for autonomous agents to interact with StarCraft II, getting observations and performing actions. Unlike the PySC2, SMAC concentrates on decentralised micromanagement scenarios, where each unit of the game is controlled by an individual RL agent.

Please refer to the accompanying paper and blogpost for the outline of our motivation for using SMAC as a testbed for MARL research and the initial experimental results.


Together with SMAC we also release PyMARL - our PyTorch framework for MARL research, which includes implementations of several state-of-the-art algorithms, such as QMIX and COMA.

Should you have any question, please reach to or

Quick Start

Installing SMAC

You can install SMAC by using the following command:

$ pip install git+

Alternatively, you can clone the SMAC repository and then install smac with its dependencies:

$ git clone
$ pip install smac/

You may also need to upgrade pip: pip install --upgrade pip for the install to work.

Installing StarCraft II

SMAC is based on the full game of StarCraft II (versions >= 3.16.1). To install the game, follow the commands bellow.


Please use the Blizzard's repository to download the Linux version of StarCraft II. By default, the game is expected to be in ~/StarCraftII/ directory. This can be changed by setting the environment variable SC2PATH.


Please install StarCraft II from The free Starter Edition also works. PySC2 will find the latest binary should you use the default install location. Otherwise, similar to the Linux version, you would need to set the SC2PATH environment variable with the correct location of the game.

SMAC maps

SMAC is composed of many combat scenarios with pre-configured maps. Before SMAC can be used, these maps need to be downloaded into the Maps directory of StarCraft II.

Download the SMAC Maps and extract them to your $SC2PATH/Maps directory. If you installed SMAC via git, simply copy the SMAC_Maps directory from smac/env/starcraft2/maps/ into $SC2PATH/Maps directory.

List the maps

To see the list of SMAC maps, together with the number of ally and enemy units and episode limit, run:

$ python -m smac.bin.map_list 

Testing SMAC

Please run the following command to make sure that smac and its maps are properly installed.

$ python -m smac.examples.random_agents

Watch a replay

You can watch saved replays by running:

$ python -m --norender --rgb_minimap_size 0 --replay <path-to-replay>

This works for any replay as long as the map can be found by the game.

For more information, please refer to PySC2 documentation.


For the detailed description of the environment, read the SMAC documentation.

The initial results of our experiments using SMAC can be found in the accompanying paper.

Citing SMAC

If you use SMAC in your research, please cite the SMAC paper.

M. Samvelyan, T. Rashid, C. Schroeder de Witt, G. Farquhar, N. Nardelli, T.G.J. Rudner, C.-M. Hung, P.H.S. Torr, J. Foerster, S. Whiteson. The StarCraft Multi-Agent Challenge, CoRR abs/1902.04043, 2019.

In BibTeX format:

  title = {{The} {StarCraft} {Multi}-{Agent} {Challenge}},
  author = {Mikayel Samvelyan and Tabish Rashid and Christian Schroeder de Witt and Gregory Farquhar and Nantas Nardelli and Tim G. J. Rudner and Chia-Man Hung and Philiph H. S. Torr and Jakob Foerster and Shimon Whiteson},
  journal = {CoRR},
  volume = {abs/1902.04043},
  year = {2019},

Code Example

Below is a small code example which illustrates how SMAC can be used. Here, individual agents execute random policies after receiving the observations and global state from the environment.

If you want to try the state-of-the-art algorithms (such as QMIX and COMA) on SMAC, make use of PyMARL - our framework for MARL research.

from smac.env import StarCraft2Env
import numpy as np

def main():
    env = StarCraft2Env(map_name="8m")
    env_info = env.get_env_info()

    n_actions = env_info["n_actions"]
    n_agents = env_info["n_agents"]

    n_episodes = 10

    for e in range(n_episodes):
        terminated = False
        episode_reward = 0

        while not terminated:
            obs = env.get_obs()
            state = env.get_state()

            actions = []
            for agent_id in range(n_agents):
                avail_actions = env.get_avail_agent_actions(agent_id)
                avail_actions_ind = np.nonzero(avail_actions)[0]
                action = np.random.choice(avail_actions_ind)

            reward, terminated, _ = env.step(actions)
            episode_reward += reward

        print("Total reward in episode {} = {}".format(e, episode_reward))


RLlib Examples

You can also run SMAC environments in RLlib, which includes scalable algorithms such as PPO and IMPALA. Check out the example code.

Latest Releases
Beta release
 Jan. 15 2019