Udacity-Self-Driving-Car

CNN based Behavioral Cloning of Self-driving Car in Udacity’s Unity Simulation. Trained the model using the Keras API.

Getting Started

Clone the repository to your local directory

git clone https://www.github.com/dipampatel18/Udacity-Self-Driving-Car.git

Install the Dependencies

Run the following command to install the dependencies

sudo pip3 install -r requirements.txt

Download the Simulator

Download

Training

To train your own model, first, data collection is required. If you wish to collect your own data, delete all the files from the folder named data inside the src folder. Now, run the simulator from the beta simulator using the following command-

Make it executable

chmod a+x beta_simulator.x86_64

Running the application

./beta_simulator.x86_64

• The simulator will start running
• Select the desired scene and start the training mode
• Press r to start the recording and navigate to the Udacity-Self-Driving-Car/src/data folder
• Press r again and start driving the car around
• The simulator is continuously recording center, left and right images, along with the steering angle, speed, throttle and brake
• All the collected data will be stored in the data folder with the images in IMG folder and the data in driving_log.csv format

Begin training using the following command

python3 train.py

After every epoch, the trained model would be saved in the same directory in the format model-00#.h5. Rename the latest (or preferable) model as model.h5 for ease of use.

Testing

The trained model has its weights saved in the model.h5 file which can be directly loaded during the testing phase

Now, just like before, launch the simulation. Only this time, select the Autonomous Mode and then run the following command to start the testing of our trained model.

python3 test.py model.h5

Saving Test Run

In order to save the FPV frames of the agent and later convert it to a video, run the following command with the desired name of the folder-

python3 test.py model.h5 test_run

To convert those frames into a video, run the following command-

python3 video.py test_run

This will save a video in the current directory with the name test_run.mp4 at 60 fps. To change the fps, feed the above command with an argment-

python3 video.py test_run --fps 30

References

• Udacity Self-Driving Car

• Udacity Self-Driving Car Simulator

• Tutorial - Siraj Raval

• Tutorial - Naoki Shibuya