Udacity Self-Driving Car Simulator

Instructions

1. Clone this repository git clone https://github.com/white43/ISY503-behavioral-cloning.git
2. cd ISY503-behavioral-cloning
3. Download datasets (see below)
4. Unpack Track-1.zip and Track-2.zip to Track-1 and Track-2 directories
5. Basic check that everything is correct so far: ls -l Track-{1,2}/f1/driving_log.csv should list two files
6. pip install -r requirements.txt

Datasets

If you're a human, remove 32-character-long random string and one dot from the file name and download it.

1. Track 1. https://torrens-files.s3.ap-southeast-2.amazonaws.com/ISY503/Track-1.94tpDvQ9gK6I5jMuX3lA2z5uG4Co4kIE.zip
2. Track 2. https://torrens-files.s3.ap-southeast-2.amazonaws.com/ISY503/Track-2.Wf9ePu1A8tK3Nn2j1Y17JzUVH3cNqqh8.zip

Training

Run the following command to train your model to drive on Track 1 (needs about 4GB of free memory):

python model.py --sources Track-1/f1 Track-2/b1 Track-1/fa1 Track-2/ba1 --train-on-autonomous-center

And the following command is for Track 2 (requires about 10GB of free memory):

python model.py --sources Track-2/f1 Track-2/f2 Track-2/fa1 Track-2/b1 Track-2/b2 Track-2/ba1 --train-on-autonomous-center

Driving

python drive.py --file model.h5

Gathering autonomous data for training

1. Create a new folder for the data to be gathered
2. Start driving with a special command line argument --save-image-to

python drive.py --file model.h5 --save-image-to path/to/directory

Training on gathered autonomous data

Autonomous data contains only center images. So, it is mandatory to provide some human-gathered data with left and right images to instruct the car how to get back when it diverges from the center of the road.

python model.py --sources [Track-N/f1, ...] Track-N/fa1 Track-N/ba1 --train-on-autonomous-center