Simple_AI_car

lecture : Intelligent control

[2022] This repo is Simple_AI_car project. Using raspberry pi 4, the goal is to complete the car along a given track.

• track

Required Library

• python 3.6.13
• tensorflow 2.3.0
• tensorflow-gpu 2.3.0

Dataset

• Download

Training data sample

• Going straight

• Truning right

• Truning left

Issue & Solution

Original model is Nvidia model. I modified this model to reduce size(parameter). It is 56 times less than the original model. I expected this modification to improve the response speed of the model. However, Reduction of model size resulted in a decrease in generalization capability. So there was a problem that it didn't work where the lighting was weak. For normal operation, I found that using multi-threads rather than speed improvement through the size of the model is the correct solution. In addition, creating low-light image data augmatation will greatly help improve performance.