Behavioral Cloning

---

Behavioral Cloning Project

The goals / steps of this project are the following:

• Use the simulator to collect data of good driving behavior
• Build, a convolution neural network in Keras that predicts steering angles from images
• Train and validate the model with a training and validation set
• Test that the model successfully drives around track one without leaving the road
• Summarize the results with a written report

Rubric Points

Here I will consider the rubric points individually and describe how I addressed each point in my implementation.

---

Files Submitted & Code Quality

1. Submission includes all required files and can be used to run the simulator in autonomous mode

My project includes the following files:

• model.py containing the script to create and train the model
• drive.py for driving the car in autonomous mode
• processor.py containing the script for loading and processing data
• model.h5 containing a trained convolution neural network
• README.md summarizing the results

2. Submission includes functional code

Using the Udacity provided simulator and my drive.py file, the car can be driven autonomously around the track by executing

python drive.py model.h5

3. Submission code is usable and readable

The model.py file contains the code for training and saving the convolution neural network. The file shows the pipeline I used for training and validating the model, and it contains comments to explain how the code works.

Model Architecture and Training Strategy

1. Model Architecture

I use Nvidia's deep learning model for self driving describe in this post. It has 3 5x5 convolution layer with stride of 2x2, followed by 2 3x3 convolution layer with stride of 1x1. After flattening, it has three feed forward layers of size 100, 50, 10 before predicting the final output. Here is a picture from the blog post:

The implementation is in model.py.

2. Model Details

I use Exponential Linear Unit (ELU) as the activation function and added dropout with keep rate of 0.5 for each feed forward layer to reduce overfitting. I use Adam optimizer without manually tuning the learning rate. The model is trained for 15 epochs.

3. Data Collection and Processing

The data set contains example data downloaded from Udacity, as well as data generated by driving 5 rounds on track1 in simulator. I pre-processed the data by cropping images , grayscale, and normalized it. Here is an example before and after pre-processing (without normalize):

In addition, I add images from left camera with +0.2 on steering angle, and images from right camera with -0.2 on steering angle.

All images are randomly shuffled and splitted into a training set containing 80% of the data, and a validation set containing 20% of the data.

The implementation is in processor.py.

4. Results

After 15 epochs, the model has 0.0262 training loss and 0.0235 validation loss.