| title | wildfire-containment-v0 |
|---|---|
| emoji | 🔥 |
| colorFrom | red |
| colorTo | yellow |
| sdk | docker |
| app_port | 7860 |
| pinned | false |
WildfireContainment-v0 is a comprehensive, real-world OpenEnv environment where an AI agent controls suppression resources to minimize fire spread and protect critical structures.
Designed for the India's First OpenEnv AI Hackathon (Meta x Hugging Face x PyTorch), it balances complex environmental dynamics (wind, spot fires, fuel consumption) with strategic resource management.
- Grid Size: 20x20
- Channels:
fuel: Available combustible material (0.0-1.0).fire_intensity: Current combustion level (0.0-1.0).moisture: Applied suppression or natural wetness (0.0-1.0).structures: Binary grid of high-value assets requiring protection.
- Dynamics:
- Spread: Fire moves to adjacent 8 cells based on ignition probability, biased by Wind Direction and Speed.
- Spot Fires: High-intensity fires have a chance to throw sparks downwind, starting new fires several cells away.
- Agents:
- 1 Air Tanker: High mobility, drops 3x3 water (moisture +0.5). Limited water capacity, must refill at grid edges.
- 2 Ground Crews: Low mobility, suppresses current 1x1 cell fire (intensity -0.2). Consumes stamina, must rest at grid edges.
A WildfireAction object containing a list of 3 unit actions:
move: 0-7 directions, 8=stay.act: Binary flag to drop water (Tanker) or suppress fire (Crew).
A WildfireObservation object containing:
- 20x20 fuel, fire, moisture, and structure grids.
- Unit locations and resource levels.
- Current wind vector.
(+)Fire intensity reduction (Fire extinguished).(+)Structure survival bonus (per step).(-)Fire spread (New cells ignited).(-)Structure loss (Large penalty).(+)Full containment bonus (Completion).
The API server must be running to process environment dynamics.
pip install -r requirements.txt
python -m uvicorn server.app:app --host 0.0.0.0 --port 7860Open a second terminal and run the following to serve the beautiful graphical UI:
cd frontend
python -m http.server 3000Then, open http://localhost:3000 in your web browser. Click Reset and Auto-Play Agent to watch the simulation!
docker build -t wildfire-env .
docker run -p 7860:7860 wildfire-envGET /tasks: Lists Easy/Medium/Hard scenarios.GET /grader: Returns current session score (0.0-1.0).POST /baseline: Run greedy heuristic baseline.