The goal of this project is to train a neural network for driving a car in the Udacity self-driving simulator and help it keep on the road.

• Prerequisites
• What is in each file?
• Training the model
• Testing the model
  • No traffic
  • Dynamic traffic
• Further work

• Python 3
• Tensorflow
• OpenCV

• Contains the code to load and process the data from the simulator.

• Functions to make the convolutional and fully connected layers.
• A network function which defines the neural network architecture.

• This is the main file that is used to train the model and save it.

The data file created by the Udacity simulator should be saved in the same directory as the python files, as in the repo. All the images should be saved in the images folder. The data file should contain either the relative or absolute paths to the images. The command to be used to train the model is as follows python3 Train.py --datafile <path_to_datafile> --save_dir <dir_to_save_tf_model> --summary_dir <dir_to_save_tf_summary> Some optional arguments with the default values in the parenthesis-

• --num_epochs the number of epochs (200)
• --minibatch_size size of the minibatch to be used (128)
• --log_file path to an existing or new file - a log of training is written in this file (log.txt).

Download the Udacity simulator Version 2 and make the file executable:

sudo chmod +x beta_simulator.x86_64

Run the simulator, choose the Jungle track and Autonomous Mode.
Run the pre-trained model:

python3 test_h5.py model.h5

Download the Term 3 simulator from the latest release and make the file executable:

sudo chmod +x term3_sim.x86_64

Install dependencies:

cd test
sh install-ubuntu.sh

The testing code is present in main.cpp
From the main directory, follow these commands to compile the code:

cd test/build
cmake .. && make

Start the Udacity simulator by running term3_sim.x86_64 Run the testing code from inside the build folder:

./path_planning

• Switch to a better simulator, with traffic and realistic conditions.
• Use Reinforcement Learning instead of Supervised Learning.