Self Driving Car Experiment

Author: (EJ) Vivek Pandey

The VR environment involved in the training/test has not been committed due to the huge file sizes and my git lfs is full lol

Dependencies

You need to have anaconda in your system first.

TensorFlow without GPU

conda env create -f environments.yml

TensorFlow with GPU

conda env create -f environment-gpu.yml

To activate the environment

source activate self-driving-car-env

To see how it works

• Clone this repo, duh!
• Get the VR env from Udacity's repo - the binary works out-of-the-box: https://github.com/udacity/self-driving-car-sim
• Launch the VR env in training mode
• Start the recording, provide a location to store the frames, and drive a minimum of 5-7 laps
• Train the model by running python model.py (Check out the arguments in the file and provide as necessary, especially the location for training images)
• Should take about 8-9 hours if you have a fairly powerful system and your training is based off of only a few laps, or else run it in a GPU instance
• If the training time scares you and you choose to use the pretrained model, feel free to skip the training steps above.
• Launch the VR env in autonomous mode
• Kick off the driver using python drive.py "model.h5" "folder_to_save_images_to"
• Watch the magic happen