The goal of this project module called PREN, is to let a robot drive through an obstacle course. There are some details that I might upload once the project is done and I get a positive response from our school to open source this. We are a team of 7 members and my main part of this project is to recognize different digits on a wall while the robot is driving. In addition, it is necessary to recognize if a crosslight is green or red. To make this work I choose to stick with deep learning to be specific CNN's. The project was first built on TensorFlow and tflearn. After a while, I had performance issues to run this on a Raspberry Pi. To trim down the parameters I needed more control over the network. I switched from tflearn to keras and could actually increase the performance by a magnitude of 10.