Skip to content

Latest commit

 

History

History
84 lines (44 loc) · 2.53 KB

README.md

File metadata and controls

84 lines (44 loc) · 2.53 KB

ROSHAN

ROSHAN (Rescue Oriented Simulation: Handling and Navigating Fires), is a wildfire simulation tool. ROSHAN integrates the principles of cellular automata with reinforcement learning to simulate wildfire dynamics and automatic handling. entral to this simulation tool is its rendering and simulation of fire spread, achieved by integrating data from the CORINE database. The interactive graphical interface of ROSHAN facilitates real-time monitoring and manipulation of fire scenarios. A key component in ROSHAN is the incorporation of a Reinforcement Learning agent, embodied as a drone, which learns to detect and mitigate fires.

You can read everything about the development here.

Agent Demo

Installation

Clone the repositories with submodules

git clone --recurse-submodules https://github.com/RoblabWh/ROSHAN.git

NodeJS

cd openstreetmap

npm install express body-parser

npm install --save-dev nodemon

CORINE CLC+

ROSHAN offers the possibility to generate custom Maps from realworld data using the CORINE Landcover Database. If you wish NOT to use custom maps you can use the small sample maps provided by this repository, you don't need to download the database in this case!

To load custom maps from all over Europe you need to register to EU Login, the European Commission's user authentication service and download the CLC+ Backbone Database:

Download Corine CLC+ Backbone - 10 meter (Year 2018 or 2021)

Once compiled you can set the dataset_path in the config.json, the standard path is /ROSHAN/assets/dataset/CLMS_CLCplus_RASTER_2021_010m_eu_03035_V1_1.tif.

Dependencies

GDAL and GDAL C++ headers

sudo apt install libgdal-dev gdal-bin libsdl2-image-dev

SDL2 - min. 2.0.17

Install SDL2 according to:

sudo apt-get install libsdl2-2.0-0 libsdl2-image-2.0-0

Anaconda & PyTorch

conda create --name roshan python=3.9 libffi==3.3

conda activate roshan

pip install torch torchvision

conda install tensorboard

conda install packaging

LLM Support

pip install transformers[torch] onnxruntime bitsandbytes optimum onnx

Compile

cd \ROSHAN

mkdir build && cd build

cmake .. && make -j&(nproc)

Usage

Start ROSHAN either as pure C++ Simulation or with Reinforcement Learning Support:

ROSHAN Sim

cd to/your/build/directory

./ROSHAN

ROSHAN Sim + Reinforcement Learning

cd ROSHAN/src/pysim/

python main.py