Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/adhiiisetiawan/q-learning-grid-env
Implementation of Q-Learning agent in grid world environment
https://github.com/adhiiisetiawan/q-learning-grid-env
gymnasium q-learning reinforcement-learning
Last synced: 3 days ago
JSON representation
Implementation of Q-Learning agent in grid world environment
- Host: GitHub
- URL: https://github.com/adhiiisetiawan/q-learning-grid-env
- Owner: adhiiisetiawan
- License: mit
- Created: 2023-08-14T13:37:49.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2023-08-15T13:18:27.000Z (over 1 year ago)
- Last Synced: 2024-01-26T06:42:57.113Z (10 months ago)
- Topics: gymnasium, q-learning, reinforcement-learning
- Language: Python
- Homepage:
- Size: 447 KB
- Stars: 1
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Q-Learning in Grid World Environments
This project focuses on implementing and demonstrating the Q-learning algorithm in two classic grid world environments: FrozenLake and TaxiV3. Q-learning is a fundamental reinforcement learning algorithm that allows an agent to learn optimal actions through exploration and exploitation.
## Table of Contents
- [Introduction to Q-Learning](#introduction-to-q-learning)
- [Grid World Environments](#grid-world-environments)
- [Implementation](#implementation)
- [Usage](#usage)
- [Results](#results)
- [License](#license)
## Introduction to Q-Learning
Q-Learning is a model-free, off-policy reinforcement learning algorithm used to find the optimal action-selection policy for a given finite Markov decision process (MDP). It's one of the foundational techniques in reinforcement learning and has been successfully applied to a wide range of problems. Q-learning aims to learn a Q-value function that represents the expected cumulative reward an agent can obtain by taking a certain action in a given state.
## Grid World Environments
This repository includes implementations for two classic grid world environments:
1. **FrozenLake**: A simple grid world where the agent must navigate across a frozen lake while avoiding holes. The agent receives a reward of +1 for reaching the goal state and a reward of 0 otherwise.
2. **TaxiV3**: An environment where the agent controls a taxi navigating through a city grid to pick up and drop off passengers at designated locations. The agent receives rewards based on its actions and progress towards delivering passengers.
## Implementation
The Q-learning algorithm has been implemented in Python using the OpenAI Gym library to interact with the grid world environments. The main components of the implementation include:
Initialization of Q-values for all state-action pairs.
Exploration and exploitation strategy (e.g., epsilon-greedy policy).
Q-value update based on the Bellman equation.
Training loop that allows the agent to interact with the environment and improve its Q-values.
## Usage
1. Clone this repository and install dependencies in your python environment
```bash
# clone repository
git clone https://github.com/adhiiisetiawan/q-learning-grid-env.git
cd q-learning-grid-env
# install dependencies
pip3 install -r requirements.txt
```
2. Start training
For frozenlake environment
```bash
python3 train.py --env FrozenLake-v1 --config config/frozenlake.yml
```
For taxi environment
```bash
python3 train.py --env Taxi-v3 --config config/taxi.yml
```
## Results
FrozenLake environment results
Taxi environment results
## License
This project is licensed under the [MIT License](https://github.com/adhiiisetiawan/q-learning-grid-env/blob/main/LICENSE).