Behavioral Cloning
Resubmission:
The file (reload_model.py) uses the trained model (model.h5). Then new images were recorded using a "recovery action mode", which means: position the car when it's ready to go off-road and start recording new images while steering towards the center. This was done approx. 5-6 times to collect enough images. Next the trained model was re-loaded (reload_model.py line 80) and the network was trained ONLY on the newly collected images from the "recovery action mode". This resulted in much better overall autonomous driving of the car. It was observed that the car stayed in the center of the road. No tire left the drivable portion of the track surface. The car never goes beyond the yellow lanes. The result of this can be viewed in the video (retrain_video.mp4).
Introduction:
This approach uses applied deep neural networks and convolutional neural networks (CNNs) to clone driving behavior by training, validating and testing a model using Keras. The model will output a steering angle to an autonomous vehicle. Data collection was accomplished using a simulator which allows for steering a car around a track. Using the collected image data and steering angles, a neural network shall be trained on the data. Finally this model shall be applied to drive the car autonomously around the track which shall serve as a validation of the neural network.
Behavioral Cloning
The goals / steps of this project are the following:
- Use the simulator to collect data characteristic 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
Files Submitted & Code Quality
1. Submission includes all required files and can be used to run the simulator in autonomous mode
Project files:
- model.py containing the script to create and train the model
- drive.py for driving the car in autonomous mode
- model.h5 containing a trained convolution neural network
- writeup_report.md or writeup_report.pdf 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.h53. 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. An appropriate model architecture has been employed
Trained on a AWS GPU (4G), my model consists of a convolution neural network with 5 convolutional layers, and filter sizes of 3x3 and 2X2 respectively. Network depths range between 24 and 64 (model.py lines 81-85)
The model includes RELU layers to introduce nonlinearity (model.py lines 81-85), and the data was normalized in the model using a Keras lambda layer (model.py line 79).
____________________________________________________________________________________________________
Layer (type) Output Shape Param # Connected to
====================================================================================================
lambda_1 (Lambda) (None, 160, 320, 3) 0 lambda_input_1[0][0]
____________________________________________________________________________________________________
cropping2d_1 (Cropping2D) (None, 90, 320, 3) 0 lambda_1[0][0]
____________________________________________________________________________________________________
convolution2d_1 (Convolution2D) (None, 43, 158, 24) 1824 cropping2d_1[0][0]
____________________________________________________________________________________________________
convolution2d_2 (Convolution2D) (None, 20, 77, 36) 21636 convolution2d_1[0][0]
____________________________________________________________________________________________________
convolution2d_3 (Convolution2D) (None, 8, 37, 48) 43248 convolution2d_2[0][0]
____________________________________________________________________________________________________
convolution2d_4 (Convolution2D) (None, 6, 35, 64) 27712 convolution2d_3[0][0]
____________________________________________________________________________________________________
convolution2d_5 (Convolution2D) (None, 4, 33, 64) 36928 convolution2d_4[0][0]
____________________________________________________________________________________________________
flatten_1 (Flatten) (None, 8448) 0 convolution2d_5[0][0]
____________________________________________________________________________________________________
dense_1 (Dense) (None, 100) 844900 flatten_1[0][0]
____________________________________________________________________________________________________
dropout_1 (Dropout) (None, 100) 0 dense_1[0][0]
____________________________________________________________________________________________________
dense_2 (Dense) (None, 50) 5050 dropout_1[0][0]
____________________________________________________________________________________________________
dropout_2 (Dropout) (None, 50) 0 dense_2[0][0]
____________________________________________________________________________________________________
dense_3 (Dense) (None, 10) 510 dropout_2[0][0]
____________________________________________________________________________________________________
dropout_3 (Dropout) (None, 10) 0 dense_3[0][0]
____________________________________________________________________________________________________
dense_4 (Dense) (None, 1) 11 dropout_3[0][0]
====================================================================================================
2. Attempts to reduce overfitting in the model
The model contains dropout layers in order to reduce overfitting (model.py lines 88,90,92).
The model was trained and validated on different data sets to ensure that the model was not overfitting (model.py line 67-68). The model was tested by running it through the simulator and ensuring that the vehicle could stay on the track. Data collection was time-consuming and cumbersome as it involved several trials using the provided simulator.
3. Model parameter tuning
The model used an Adam Optimizer, so the learning rate did not have to be tuned manually (model.py line 96).
4. Appropriate training data
Training data was chosen to keep the vehicle driving on the road. I used a combination of center lane driving.
The data collection strategy I used was as follows:
i. 3 laps forward direction
ii. 3 laps reverse direction
iii. 2 laps steering right to left and vice versa.
For details about how I created the training data, see the next section.
Model Architecture and Training Strategy
1. Solution Design Approach
The overall strategy for deriving a model architecture was to seek a well established CNN and repurpose it for behavioral cloning.
I used a convolution neural network model which is similar to (http://images.nvidia.com/content/tegra/automotive/images/2016/solutions/pdf/end-to-end-dl-using-px.pdf)
I thought this model might be appropriate because it is well known and has been tested in the real-world scenario.
In order to gauge how well the model was working, I split my image and steering angle data into a training and validation set. I found that my first model had a low mean squared error on the training set but a high mean squared error on the validation set. This implied that the model was overfitting.
To combat the overfitting, I modified the model so that it has dropout layers (model.py lines: 88,90,92)
The final step was to run the simulator to see how well the car was driving around track one. I found that the result was very good based on the data I personally collected (see video output file) and my model.h5 The car successfully completed the entire first lap without going off the road and always staying in-between the lane markers. The car accomplished the goal of being able to drive autonomously around the track without leaving the road.
2. Final Model Architecture
The final model architecture (model.py lines 78-93) consisted of a convolution neural network with the following layers and layer sizes: 5 CNNs, 5x5, 3x3 layers.
3. Creation of the Training Set & Training Process
To capture good driving behavior, I first recorded 3 laps on track one using center lane driving. Here is an example image of center lane driving:
Then I repeated this process on by recording 3 laps in the reverse direction. Finally I recorded another 2 laps where the vehicle recovering from the left side and right sides of the road back to center so that the vehicle would learn to handle these driving behaviors.
To augment the data set I used Data Augmentation which helps the model generalize better but also expands
the training set. Using this technique I flipped images and angles thinking that this would help with turn bias.
For example, here is an image that has then been flipped using the OpenCV 2 library. By doing so you can flip and invert
the steering angle clockwise and counter-clockwise allowing the car to drive better.
After the collection process, I ended with a CSV file just under 5MB. I managed to collect a total of 12,349 data points. I then preprocessed this data by using model.py
I finally randomly shuffled the data set and put 25% of the collected data points into a validation set.
I used this training data for training the model. The validation set helped determine if the model was over or under fitting. I found that the ideal number of epochs was 5 and I used batch size of 32.