generals 0.1.0

Generals.io environment compliant with PettingZoo API standard powered by Numpy.

Generals.io RL

Installation • Customization • Environment • Getting Started

Generals.io is a real-time strategy game where players compete to conquer their opponents' generals on a 2D grid. While the goal is simple — capture the enemy general — the gameplay involves a lot of depth. Players need to employ strategic planning, deception, and manage both micro and macro mechanics throughout the game. The combination of these elements makes the game highly engaging and complex.

This repository aims to make bot development more accessible, especially for Machine Learning based agents.

Highlights:

• 🚀 Fast & Lightweight simulator powered by numpy (thousands of steps per second)
• 🦁 Compatibility with Reinforcement-Learning API standards Gymnasium and PettingZoo
• 🔧 Easy customization of environments
• 🔬 Analysis tools such as replays

Generals.io has several interesting properties:

• 👀 Partial observability
• 🏃‍♂️ Long action sequences and large action spaces
• 🧠 Requires strategical planning
• ⏱️ Real-time gameplay

📦 Installation

Stable release version is available through pip:

pip install generals

Alternatively, you can install latest version via git

git clone https://github.com/strakam/Generals-RL
cd Generals-RL
pip install -e .

Usage Example (🦁 PettingZoo)

from generals.env import pz_generals
from generals.agents import RandomAgent
from generals.config import GameConfig

# Initialize agents - their names are then called for actions
agents = {
    "Red": RandomAgent("Red"),
    "Blue": RandomAgent("Blue")
}

game_config = GameConfig(
    grid_size=16,
    mountain_density=0.2,
    city_density=0.05,
    general_positions=[(2, 12), (8, 9)],
    agent_names=list(agents.keys()),
)

# Create environment
env = pz_generals(game_config, render_mode="human") # render_mode {"none", "human"}
observations, info = env.reset(options={"replay_file": "test"})

# How fast we want rendering to be
actions_per_second = 2

while not env.game.is_done():
    actions = {}
    for agent in env.agents:
        # Ask agent for action
        actions[agent] = agents[agent].play(observations[agent])
    # All agents perform their actions
    observations, rewards, terminated, truncated, info = env.step(actions)
    env.render(tick_rate=actions_per_second)

Usage example (🤸 Gymnasium)

from generals.env import gym_generals
from generals.agents import RandomAgent
from generals.config import GameConfig

# Initialize agent
agent = RandomAgent("Red")

game_config = GameConfig(
    grid_size=16,
    mountain_density=0.2,
    city_density=0.05,
    general_positions=[(2, 12), (8, 9)],
    agent_names=[agent.name]
)

# Create environment
env = gym_generals(game_config, render_mode="human") # render_mode {"none", "human"}
observation, info = env.reset(options={"replay_file": "test"})

# How fast we want rendering to be
actions_per_second = 2
done = False

while not done:
    action = agent.play(observation)
    observation, reward, terminated, truncated, info = env.step(action)
    done = terminated or truncated
    env.render(tick_rate=actions_per_second)

🎨 Customization

The environment can be customized via GridConfig class or by creating a custom map.

🗺️ Random maps

from generals.env import pz_generals
from generals.config import GameConfig

game_config = GameConfig(
    grid_size=16,                          # Edge length of the square grid
    mountain_density=0.2,                  # Probability of a mountain in a cell
    city_density=0.05,                     # Probability of a city in a cell
    general_positions=[(0,3),(5,7)],       # Positions of generals (i, j)
    agent_names=['Human.exe','Agent007']   # Names of the agents that will be called to play the game
)

# Create environment
env = pz_generals(game_config, render_mode="none")
observations, info = env.reset()

🗺️ Custom maps

Maps can be described by strings. We can either load them directly from a string or from a file.

from generals.env import pz_generals
from generals.config import GameConfig

game_config = GameConfig(
    agent_names=['Human.exe','Agent007']  # Names of the agents that will be called to play the game
)
map = """
.3.#
#..A
#..#
.#.B
"""

env = pz_generals(game_config, render_mode="none")
env.reset(map=map) # Here map related settings from game_config are overridden

Maps are encoded using these symbols:

• . for passable terrain
• # for non-passable terrain
• A,B are positions of generals
• digits 0-9 represent cost of cities calculated as (40 + digit)

🔬 Replay Analysis

We can store replays and then analyze them.

Storing a replay

from generals.env import pz_generals
from generals.config import GameConfig

game_config = GameConfig()
options = {"replay_file": "replay_001"}
env = pz_generals(game_config, render_mode="none")
env.reset(options=options) # encodes the next game into a "replay_001" file

Loading a replay

The following code loads and executes replay named replay_001:

import generals.utils

generals.utils.run_replay("replay_001")

🕹️ Replay controls

• q — quit/close the replay
• ←/→ — increase/decrease the replay speed
• h/l — move backward/forward by one frame in the replay
• spacebar — toggle play/pause
• mouse click on the player's row — toggle the FoV (Field Of View) of the given player

🌍 Environment

🔭 Observation

An observation for one player is a dictionary of 8 key/value pairs. Each value is a 2D np.array containing information for each cell. Values are (binary) masked so that only information about cells that an agent can see can be non-zero.

Key	Shape	Description
army	(N,N,1)	Number of units in a cell regardless of owner
general	(N,N,1)	Mask of cells that are visible to the agent
city	(N,N,1)	Mask saying whether a city is in a cell
visibility	(N,N,1)	Mask indicating cells that are visible to the agent
ownership	(N,N,1)	Mask indicating cells controlled by the agent
ownership_opponent	(N,N,1)	Mask indicating cells owned by the opponent
ownership_neutral	(N,N,1)	Mask indicating cells that are not owned by agents
structure	(N,N,1)	Mask indicating whether cells contain cities or mountains, even out of FoV
action_mask	(N,N,4)	Mask where [i,j,k] indicates whether you can move from a cell [i,j] to direction k where directions are in order (UP, DOWN, LEFT, RIGHT)
n_land	(1,)	Int representing number of cells an agent owns
n_army	(1,)	Int representing total number of units of an agent over all cells
is_winner	(1,)	Bool representing whether an agent won
timestep	(1,)	Timestep

ℹ️ Information

The environment also returns information dictionary for each agent, but it is the same for everyone. This might potentially contain debug information.

Key	Type	Description
army	Int	Total number of units that the agent controls
land	Int	Total number of cells that the agent controls
is_winner	Bool	Boolean indicator saying whether agent won

Example:

print(info['red_agent']['army'])

⚡ Action

Action is an np.array([i,j,k,h]) indicating that you want to move units from cell [i,j] in a direction k.

The last value h signals whether to send 1 (half) of units or 0 (all) units to the neighboring cell.

🎁 Reward

It is possible to implement custom reward function. The default is 1 for winner and -1 for loser, otherwise 0.

def custom_reward_fn(observation, info):
    # Give agent a reward based on the number of cells they own
    return {
        agent: info[agent]["land"]
        for agent in observation.keys()
    }

env = generals_v0(reward_fn=custom_reward_fn)
observations, info = env.reset()

🚀 Getting Started

Creating your first agent is very simple. Start by subclassing an Agent class just like RandomAgent here.

• Every agent must have a name as it is his ID by which he is called for actions.
• Every agent must implement play(observation) function that takes in observation and returns an action as described above.
• You can simply follow examples to make your bot running.
• When creating an environment, you can choose out of two render_modes:
  • none that omits rendering and is suitable for training,
  • human where you can see the game roll out.

🛠️ Coming Soon

• Extend action space to sending half of units to another square
• Examples and baselines using RL
• Add human control to play against
• New analysis tools

Requests for useful features and additions are welcome 🤗.