RAD-TEAM

Multi-Agent Deep Reinforcement Learning for Radiation Localization Radiation Localization by Ben Totten

Citation

If you find this code helpful in your research, please cite:

@mastersthesis{totten2023radteam,
  title        = {Multi-Agent Deep Reinforcement Learning for Radiation Localization Radiation Localization},
  author       = {Benjamin Totten},
  year         = 2023,
  month        = {August},
  address      = {Portland, OR},
  note         = {Available at \url{[https://example.com/thesis.pdf](https://pdxscholar.library.pdx.edu/cgi/viewcontent.cgi?article=7590&context=open_access_etds)}},
  school       = {Portland State University},
  type         = {Masters thesis}
}

Notice

For code used for this Thesis, see the Masters thesis branch.

Resources

• Multi-agent architecture created by Alagha et al. and their published paper
• Single-agent architecture created by Liu et al. and their published paper
• Single-agent radiation source search environment created by Proctor et al. and their published paper.

Documentation

Documentation generated with Sphinx.

Generate documentation with sphinx-build -b html docs doc_build from root directory

Prerequisites

• [Optional] Environment manager:
  • [Recommended] Micromamba

Setup

1. Initialize environment

   1. [Option 1] with environment manager

      micromamba create --file environment.yml

      micromamba activate rad-team

   2. [Option 2] with pip

      pip install -e .

2. Run

   launch_radteam

   python -m rad_team

** Note: The RAD-A2C implementation from Proctor et al. requires OpenMPI for parallel processing.

Training

Dont forget to activate any python environments micromamba activate rad-team

Testing

Run all tests with make test

Clean

Refresh with make clean

Docs

Generate docs with make docs. These will generate in the docs/build/html folder.

Distributed Evaluation Mode

The evaluation portion of this codebase uses Ray Clusters. Each episode runs as it's own Actor