CS221-CS229 Class project. The objective is to predict steering command from raw camera input.

src: host the main program

plots: contains some relevant illustrations

In the src folder, there are 2 subfolders:

• classification: CNN architecture to predict steering direction
• regression : 3 different CNN architectures to predict the steering angle

The dataset is provided by Udacity and contains a lot of information (3 different camera views, steering, brake and throttle reports). We used the 12 minutes part of the dataset and it can be found here: dataset

Note: when running the code be careful that your data files are in the right path.

Extract the data: Need rosbag to be installed. running src/parse_bag.py will convert the rosbag format into a csv file steering.csv containing the steering command and split the images into 3 folders (center_camera, left_camera and right_camera). The images are saved in jpeg format and the name of the images corresponds to the ROS time stamp.

Note: The time stamp of the steering command and of the images are not matching perfectly (different frequency). Some pre-processing will have to be done to map the time stamps.

• Python 2.7
• ROS : ros-kinetic distribution
• OpenCV for python
• Keras (with TensorFlow backend)
• Matlab r2016b