mitrefireline / simfire

An open-source wildfire simulator written in Python and meant to be used to train reinforcement learning (RL) agents.
https://mitrefireline.github.io/simfire/
Apache License 2.0
30 stars 2 forks source link
python reinforcement-learning reinforcement-learning-environments simulation simulator wildfire

SimFire Fire Simulator

Introduction

SimFire uses PyGame to display and simulate different fire spread models, including the Rothermel Surface fire spread model described in this paper.

For more comprehensive documentation, go to our docs page.

This repository is part of the MITRE Fireline project and is associated with both BurnMD and SimHarness. BurnMD is used in the HistoricalDataLayer to provide historical fire data for the simulation. SimHarness is a reinforcement learning training harness that uses this simulator to train agents to fight fires.

Running the Simulation


Left: Fire simulated near Julian, CA. Right: Fire simulated near Reno, NV.
Both fires have winds from the east at 20mph

Install simfire by following the installation instructions. Then run the run_game.py script:

python run_game.py

Running as a Python Module

from simfire.sim.simulation import FireSimulation
from simfire.utils.config import Config

config = Config("configs/operational_config.yml")
sim = FireSimulation(config)

# Run a 1 hour simulation
sim.run("1h")

# Run the same simulation for 30 more minutes
sim.run("30m")

# Render the next 2 hours of simulation
sim.rendering = True
sim.run("2h")

# Now save a GIF and fire spread graph from the last 2 hours of simulation
sim.save_gif()
sim.save_spread_graph()
# Saved to the location specified in the config: simulation.sf_home

# Update agents for display
# (x, y, agent_id)
agent_0 = (5, 5, 0)
agent_1 = (5, 5, 1)

agents = [agent_0, agent_1]

# Create the agents on the display
sim.update_agent_positions(agents)

# Loop through to move agents
for i in range(5):
    agent_0 = (5 + i, 5 + i, 0)
    agent_1 = (5 + i, 5 + i, 1)
    # Update the agent positions on the simulation
    sim.update_agent_positions([agent_0, agent_1])
    # Run for 1 update step
    sim.run(1)

# Turn off rendering so the display disappears and the simulation continues to run in the
# background
sim.rendering = False

Installing the Package

pip install simfire

Contributing

For contributing, see the Contribution Page in our docs.

Citation

To cite this software, use the “Cite this repository” link built into GitHub on the right.

Copyright

Copyright ©2023 The MITRE Corporation. ALL RIGHTS RESERVED. Approved for Public Release; Distribution Unlimited. Public Release Case Number 22-3261.