Auto Encoder CNN for Maze Path Finding

• leftmost column shows the mazes
• the middle column are the approximation of solution
• the rightmost column shows the true solutions

• Summary: This project, part of EE456 at Penn State, involved implementing and testing an Auto Encoder Convolutional Neural Network (CNN) to solve a maze path finding problem. The goal was to find a path from the bottom left to the top right corner of a maze represented as an image.

• Tools Used: The project extensively used Python 3.8.15 with libraries such as torch, torchvision, matplotlib, and numpy in an Anaconda environment.

File Descriptions

1. temp.ipynb: A Jupyter notebook containing the initial experiments and tests conducted during the early stages of the project. It includes exploratory data analysis and preliminary model training.

2. generateData.ipynb: This notebook is responsible for generating the maze data used for training the Auto Encoder. It outlines the process of creating 10,000 unique mazes and their corresponding solution paths.

3. AutoEncoderNet.ipynb: This file contains the complete implementation of the Auto Encoder CNN, including the architecture definition, training process, and testing on generated maze data.

4. FinalProject_EfeSahin.pdf: Comprehensive documentation of the project covering the objective, methodology, neural network architecture, results, and conclusions. It also includes visual representations of training progress and the final results.

5. EE456Final Project.mp4: A video presentation of the project, illustrating the working of the Auto Encoder CNN in solving the maze problems and showcasing the project's outcomes.

The project demonstrates the effectiveness of an Auto Encoder CNN in solving path finding problems in mazes, highlighting the capability of neural networks in feature extraction and image reconstruction.

Here are links to my other work in EE456

1. Implemented with pytorch: cifar-cnn
2. Implemented with MatLAB: mlp-backprop