Repo for Predicting Steering Wheel

TensorFlow implementation based on End to End Learning for Self-Driving Cars.

Preprocessing

input.ipynb

Model training

nvidia_steering.ipynb

Extracting camera images - Tools

./bag_to_images.py dataset.bag right_camera/ /right_camera/image_color
./bag_to_images.py dataset.bag left_camera/ /left_camera/image_color
./bag_to_images.py dataset.bag center_camera/ /center_camera/image_color

Extracting camera timestamps - Tools

./camera_timestamps.py dataset.bag timestamps-center.csv /center_camera/image_color
./camera_timestamps.py dataset.bag timestamps-left.csv /left_camera/image_color
./camera_timestamps.py dataset.bag timestamps-right.csv /right_camera/image_color

Extracting steering angles - Tools

This script can extract any topic data to csv.

./bag_to_csv.py dataset.bag steering.csv /vehicle/steering_report

Data preprocessing - Tools

Image resizing, pickling and steering interpolation is implemented in steering_input.ipynb

Data augmentation

Generating random, labeled frames from original frames:

import augmentation as aug

transformed_image, new_steering_wheel_angle, rotation, shift = aug.steer_back_distortion(
                                                                    image, 
                                                                    steering_wheel_angle, 
                                                                    speed)

Generating distorted images:

import augmentation as aug
rotation = 0.01 # radians
shift = 0.5 # meters
distorted = aug.apply_distortion(img, rotation, shift)

Model definition and training

NVIDIA_Steering.ipynb