Behavioral Cloning

by Mohit Arvind Khakharia

---

Overview

In this project, we will use traditional and convolutional neural networks to clone driving behavior. We will train, validate and test a model using Keras. The model will output a steering angle to an autonomous vehicle.

This project was done as a part of Udacity's Self-Driving Car Nanodegree Program. The model's performance has been tested on for resolution of 640x480, and graphic quality selected as 'fantastic'.

Model performance on track 1

DRIVER CAM	AERIAL CAM

---

Goals

• 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

Details About Files In This Directory

drive.py

Usage of drive.py requires you have saved the trained model as an h5 file, i.e. model.h5. See the Keras documentation for how to create this file using the following command:

model.save(filepath)

Once the model has been saved, it can be used with drive.py using this command:

python drive.py model.h5

The above command will load the trained model and use the model to make predictions on individual images in real-time and send the predicted angle back to the server via a websocket connection.

Note: There is known local system's setting issue with replacing "," with "." when using drive.py. When this happens it can make predicted steering values clipped to max/min values. If this occurs, a known fix for this is to add "export LANG=en_US.utf8" to the bashrc file.

Saving a video of the autonomous agent

python drive.py model.h5 run1

The fourth argument, run1, is the directory in which to save the images seen by the agent. If the directory already exists, it'll be overwritten.

ls run1

[2017-01-09 16:10:23 EST]  12KiB 2017_01_09_21_10_23_424.jpg
[2017-01-09 16:10:23 EST]  12KiB 2017_01_09_21_10_23_451.jpg
[2017-01-09 16:10:23 EST]  12KiB 2017_01_09_21_10_23_477.jpg
[2017-01-09 16:10:23 EST]  12KiB 2017_01_09_21_10_23_528.jpg
[2017-01-09 16:10:23 EST]  12KiB 2017_01_09_21_10_23_573.jpg
[2017-01-09 16:10:23 EST]  12KiB 2017_01_09_21_10_23_618.jpg
[2017-01-09 16:10:23 EST]  12KiB 2017_01_09_21_10_23_697.jpg
[2017-01-09 16:10:23 EST]  12KiB 2017_01_09_21_10_23_723.jpg
[2017-01-09 16:10:23 EST]  12KiB 2017_01_09_21_10_23_749.jpg
[2017-01-09 16:10:23 EST]  12KiB 2017_01_09_21_10_23_817.jpg
...

The image file name is a timestamp of when the image was seen. This information is used by video.py to create a chronological video of the agent driving.

video.py

python video.py run1

Creates a video based on images found in the run1 directory. The name of the video will be the name of the directory followed by '.mp4', so, in this case the video will be run1.mp4.

Optionally, one can specify the FPS (frames per second) of the video:

python video.py run1 --fps 48

Will run the video at 48 FPS. The default FPS is 60.

Files Submitted & Code Quality

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

• model.py contains the script to create and train the model
• drive.py drives the car in autonomous mode
• model.h5 contains a trained convolution neural network
• writeup_report.md (also in README.md) summarizes the results
• view_data.ipynb shows various aspects of the data.

2. 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. Usability and Readability

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.

Attempts to reduce overfitting in the model

• Two Dropout layers
• Early stopping
• Data Augmentation by :
  a) Using Left and Right Images with correction
  b) Image Fliping
  :-------------------------:

Model parameter tuning

Hyperparameter	Value
Learning Rate	Used Adam optimizer(learning rate was not tuned manually)
Batch Size	32
Epoch	5
Training Data Percentage	80%
Validation Data Percentage	20%
Steering Correction for Left and Right Cameras	0.2
Dropout Prob	0.5

Training data

Training data was chosen to keep the vehicle driving on the road. I used a combination of center lane driving, recovering from the left and right sides of the road

Track1
Forward Lap	3
Backward Lap	1
Recovery Areas	5

Note - The car was driven using the mouse so that we get the maximum amount of non-zero steering angle frames. The distribution of the steering angle in the training data is shown below.

Frame count vs Steering Angle

Preprocessing

• Image Normalization : The images are normalized so that the computations are neither too big or small.
• Image Cropping : The sky and the car dome dont affect the steering angle and so they were cropped.

Model Architecture and Training Strategy

1. Different Approaches tried

• Convolution neural network model similar to the Alexnet with last layer as fully connected layer with one unit.
• Convolution neural network model similar to comma.ai ([github link (https://github.com/commaai/research/blob/master/train_steering_model.py)).
• NVIDIA CNN

2. Final Model Architecture

Network architecture is modified from NVIDIA CNN is used which consists of 9 layer, including

• 1 Normalization layer
• 1 Cropping layer
• 3 convolutional layers with subsampling and Rectified Linear Unit(RELU)
• 2 convolutional layers with only Rectified Linear Unit(RELU)
• 2 Dropout layers for generalization
• 3 fully connected layers.

NVIDIA CNN

Modified NVIDIA CNN

Training Strategy

Initially because of the small dataset, the model was memorizing and therefore there was overfitting. But, after suffucient data was collected the validation dataset became better.

The final step was to run the simulator to see how well the car was driving around track one. There were a few spots where the vehicle fell off the track. To improve the driving behavior in these cases, I added the recovery data to comeback from situations where the car is drifitng on the side.

0400/30347 [==============================] - 34s - loss: 0.0049 - val_loss: 0.0031
Epoch 2/5
30358/30347 [==============================] - 32s - loss: 0.0036 - val_loss: 0.0026
Epoch 3/5
30400/30347 [==============================] - 32s - loss: 0.0037 - val_loss: 0.0033
Epoch 4/5
30358/30347 [==============================] - 32s - loss: 0.0036 - val_loss: 0.0031
Epoch 5/5
30400/30347 [==============================] - 32s - loss: 0.0035 - val_loss: 0.0027

Track 1
Training Loss	0.0035
Validation Loss	0.0027

RESULT

The vehicle is able to drive autonomously around the track without leaving the road.

Future Improvements

• More augmented data using brightness, hue and scaled images so that the CNN learns to identify the road curvature under different lighting conditions.
• Add different types of shadows onto the image frames so that the detection is shadow agnostic.
• Use CAPSULE NETWORKS so that the detection can tackle pixel attacks and maintain the positional constraints(E.g - An image of a road in the Advertisement banner on the side of the track should not effect the decision making process)

Model performance on Track 1

DRIVER CAM	AERIAL CAM