Goals
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
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
- model.h5 containing a trained convolution neural network
- writeup_report.md or writeup_report.pdf summarizing the results
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.h5The 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.
| Layer (type) | Output Shape | Param |
|---|---|---|
| conv2d_1 (Conv2D) | (None, 40, 80, 24) | 1824 |
| activation_1 (Activation) | (None, 40, 80, 24) | 0 |
| max_pooling2d_1 (MaxPooling2) | (None, 39, 79, 24) | 0 |
| conv2d_2 (Conv2D) | (None, 20, 40, 36) | 21636 |
| activation_2 (Activation) | (None, 20, 40, 36) | 0 |
| max_pooling2d_2 (MaxPooling2) | (None, 19, 39, 36) | 0 |
| conv2d_3 (Conv2D) | (None, 10, 20, 48) | 43248 |
| activation_3 (Activation) | (None, 10, 20, 48) | 0 |
| max_pooling2d_3 (MaxPooling2) | (None, 9, 19, 48) | 0 |
| conv2d_4 (Conv2D) | (None, 9, 19, 64) | 27712 |
| activation_4 (Activation) | (None, 9, 19, 64) | 0 |
| max_pooling2d_4 (MaxPooling2) | (None, 8, 18, 64) | 0 |
| conv2d_5 (Conv2D) | (None, 8, 18, 64) | 36928 |
| activation_5 (Activation) | (None, 8, 18, 64) | 0 |
| max_pooling2d_5 (MaxPooling2) | (None, 7, 17, 64) | 0 |
| flatten_1 (Flatten) | (None, 7616) | 0 |
| dense_1 (Dense) | (None, 100) | 761700 |
| activation_6 (Activation) | (None, 100) | 0 |
| dropout_1 (Dropout) | (None, 100) | 0 |
| dense_2 (Dense) | (None, 50) | 5050 |
| activation_7 (Activation) | (None, 50) | 0 |
| dropout_2 (Dropout) | (None, 50) | 0 |
| dense_3 (Dense) | (None, 10) | 510 |
| activation_8 (Activation) | (None, 10) | 0 |
| dense_4 (Dense) | (None, 1) | 11 |
| Total params: 898,619 | Trainable params: 898,619 | Non-trainable params: 0 |
The model contains dropout layers in order to reduce overfitting (model.py lines 21).
I try tune batch size for remove overfitting
- Tuned dropout layer
- learning rate
- batch size
- kernel size of filters in conv layer
Firstly I created data from the simulator and then process the csv generated by the simulator to create new csv in which I put center, left and right image. For this I adjusted steering angle for left and right image. After that I created mirror image of each image and reverse the steering angle.
Convolution neural network is best suited for image analysis, so I used CNN in first few layers of model with different kernel size and number of filters.
To avoid overftting I have add dropout layer when I'm adding dense layer in model.
Final Architecture is not changed, I kept it as older one but try tune few param like learning rate and dropout to get better result
To capture good driving data, In first drive iteration I drive normally and collected data. After first iteration I try to collect data on reverse track.
For fine-tuning data I try to collect few edge cases on a turn and try to move vehicle to center from left and right
To capture good driving behavior, I first recorded two laps on track one using center lane driving. Here is an example image of center lane driving:
Here few image from dataset where first 3 are normal and last 3 are flipped image.
