/DQN

Primary LanguagePython

Deep Q-Learning Hospital Navigation Agent

Project Overview

This project implements a reinforcement learning agent using Deep Q-Learning to navigate a simulated hospital environment. The agent, representing a nurse assistant, must efficiently navigate a 5x5 grid to reach a medicine cabinet while avoiding collisions with doctors and nurses.

Environment Description

  • 5x5 grid representing a hospital floor
  • Agent: Nurse assistant (🙂)
  • Static objects: 5 Beds, 1 Medicine Cabinet, 1 Doctor, 1 Nurse
  • Actions: Move Up, Down, Left, Right
  • Rewards:
    • Reaching Medicine Cabinet: +10
    • Reaching a Bed: +1
    • Colliding with Doctor or Nurse: -5
    • Each step: -0.1 (to encourage efficiency)
  • Termination Conditions:
    • Reaching the Medicine Cabinet (success)
    • Colliding with Doctor or Nurse (failure)
    • Max steps reached (100 steps)

Project Structure

  • HospitalNavigation_env.py: Custom Gym environment implementation
  • train.py: Script to train the DQN agent
  • play.py: Script to run simulations with the trained agent
  • hospital_navigation_weights.h5f: Saved weights of the trained model
  • requirements.txt: List of required Python packages

Installation

  1. Clone this repository:

git clone https://github.com/freedisch/DQN.git
cd DQN
  1. Install the required packages:

pip install -r requirements.txt

Usage

  1. To train the agent:

python train.py
  1. To run a simulation with the trained agent:

python play.py

Video Demonstration

Link to 5-minute video demonstration

Technologies Used

  • Python 3.8+
  • OpenAI Gym
  • Keras
  • Keras-RL
  • TensorFlow

Author

Thibaut Batale

License

This project is licensed under the MIT License - see the LICENSE file for details.