Skip to content

Latest commit

 

History

History
46 lines (31 loc) · 1.53 KB

README.md

File metadata and controls

46 lines (31 loc) · 1.53 KB
Raw

Landmark Complex Exploration via Deep Reinforcement Learning

This project aims to achieve multi-agent cooperative exploration in sparse landmark complex environments. Landmark complex is an abstract simplicial complex proposed for assisting multi-agent exploration. Although traditional landmark complex exploration methods are frontier-based, our method uses deep reinforcement learning to learn a policy for exploration and a three-stage curriculum to help mitigate reward sparsity. This Github repository includes all necessary codes to reproduce our result in Multi-Agent Exploration of an Unknown Sparse Landmark Complex via Deep Reinforcement Learning.

Requirements

  • Unity 2020.3.35
  • ML-agents Release 19
  • Python 3.8
  • Docker

Usage

  • First navigate to the Python directory and build the docker image
docker build -t lcserver .
  • Run the Docker container
docker run -d --name lcserver -p 80:80 lcserver

  • Open the root directory with Unity Editor 2020.3.35

  • In the Unity Editor, navigate to Assets/LandmarkComplex/Scenes and open LandmarkComplex_Experiment

  • Press the Play button, and the exploration will begin.

  • By pressing the Play button again, the exploration will be terminated.

  • To get the exploration data, export the container

docker export container_id > res.tar
  • The experiment data can be found at app/data_output/DRL/out.csv