This project implements a reinforcement learning (RL) environment and agents for a Trash collector robot simulation. The robot navigates a grid, collects trash, and avoids obstacles while learning optimal strategies using Q-Learning and Deep Q-Networks (DQN).
- Environment: A grid-based simulation with customizable trash and obstacle counts.
- Agents: Includes Q-Learning and DQN agents for training and decision-making.
- Rendering: Smooth and realistic agent movement using
pygame. - Reinforcement Learning: Supports exploration vs. exploitation strategies and neural network-based Q-value approximation.
- Clone the repository:
git clone https://github.com/ahmadkh1995/Trash_Collector_Robot_RL.git cd Trash_Collector_Robot_RL - Install dependencies:
pip install -r requirements.txt
- Run the Simulation
To start the simulation, execute the following command:
python main.py
Train the Agent
Modify the agent configuration in agent/DQN_learning_agent.py or agent/Q_learning_agent.py and run the simulation to train the agent.
Environment Customization
You can customize the environment in environment/garbage_collector_env.py:
Number of obstacles: Set num_obstacles in the reset method.
Number of trash items: Set num_trashes in the reset method.
project/
├── main.py # Entry point for the simulation.
├── environment/
│ └── garbage_collector_env.py # Defines the RL environment.
├── agent/
│ ├── DQN_learning_agent.py # Implements the DQN agent.
│ └── Q_learning_agent.py # Implements the Q-Learning agent.
├── assets/
│ └── icons/ # Contains icons for rendering the agent, trash, and obstacles.
└── requirements.txt $ Contains requir3ed libraries to be installed
Python 3.8+
numpy
torch
pygame
gymnasium
Environment: The robot moves in a grid, collecting trash and avoiding obstacles.
Agents: RL agents learn optimal strategies using Q-Learning or DQN.
Rendering: The simulation is visualized using pygame with smooth agent movement.
Author: Ahmad Kheirandish - GitHub: ahmadkh1995