Behavioral Cloning: Using Deep Learning to Clone Driving Behavior

Overview

---

This is Udacity's Self-Driving Car Nanodegree Project.

The goal of this project is to simulate self-driving by using behavior cloning. In other words, the "car" is trying to mimic the driver (me) without any other aid (ex: pathplanning or preset instructions). In this project, I trained my car by using recorded driving images from a simulator provided by Udacity. In the simulator, I recorded myself driving the simulated car. The data that was captured consists of images (from 3 different camera angles), steering angle, throttle, speed, and braking. I used the images from the 3 camera anngles and the steering angle to train my model. The model was then designed and trained to predict a steering angle in the future, so that it can drive the car autonomously. The model was evaluated with a track it had never seen before.

Suggested Reading:

• End to End Learning for Self-Driving Cars
• Going Deeper with Convolutions
• DeepTesla: End-to-End Learning from Human and Autopilot Driving

I used the above research papers to learn more and develop my model

Contents

• model.ipynb (notebook used to create and train the model)
• drive.py (script to drive the car - feel free to modify this file)
• model.h5 (a trained Keras model)
• a report writeup file

Dependencies

• Anaconda 4.2
• Keras
• TensorFlow
• OpenCV

Completed Product

Checkout:

Writeup.md for more details