The-GhostCar-An-Autonomous-Car-Project
COMPLETE PROJECT @The GhostCar - Autonomous Car Prototype
AIM
The Aim of this Project was to build a Fully Autonomous Car Simulation using Behavioural Cloning on Convolution Neural Networks Model.
MOTIVATION
Autonomous cars are a boon for the vision of a smart world. Tesla, BMW, Google, NVIDIA are some of the companies contributing to this technology and have overcome major milestones in developing autonomous vehicles. With the development of autonomous cars, they can cause a huge boost to the transportation and automobile industries and hence we aimed to contribute to the Autonomous Car Development Industry.
CONTENTS
1) Drive.py - The Python Program hosts a Flask server on Local host to communicate controls to the simulation softwares.
2) Behavioural Cloning.ipynb - Python Notebook File that shows the procedure and methods developed and used in order to build and train the CNN Model for the Autonomous Car.
3) Model.h5 - Weights file for several trained models.
MODELS
1) model1.h5 - Trained over 4053 samples and gives a loss of 0.0743
2) model2.h5 - Trained over 8205 samples and gives a loss of 0.369
3) model3.h5 - (Best) Trained over 30227 samples and gives a loss of 0.0276
Summary of the Design CNN Model
Note: The Model Developed is as per the Nvidia End-to-End Self Driving Car Documentation
_______________________
Layer (type) Output Shape Param #
=================================================================
conv2d_21 (Conv2D) (None, 31, 98, 24) 1824
_______________________
conv2d_22 (Conv2D) (None, 14, 47, 36) 21636
_______________________
conv2d_23 (Conv2D) (None, 5, 22, 48) 43248
_______________________
conv2d_24 (Conv2D) (None, 3, 20, 64) 27712
_______________________
conv2d_25 (Conv2D) (None, 1, 18, 64) 36928
_______________________
flatten_5 (Flatten) (None, 1152) 0
_______________________
dense_17 (Dense) (None, 100) 115300
_______________________
dense_18 (Dense) (None, 50) 5050
_______________________
dense_19 (Dense) (None, 10) 510
_______________________
dense_20 (Dense) (None, 1) 11
=================================================================
Total params: 252,219
Trainable params: 252,219