Steering Angle Control of Self Driving Car by CNN Network and Image Processing.
The Goal of our project is to define a CNN model which will generate the steering angle input that will allow a car to drive autonomously on flat road conditions. We trained our model using visual data that was generated by a car driven in Udacity simulator. The validation after training is also done on the same simulator. The udacity simulator gives us the access to test our model by using the python interface. In this simulator there are two tracks - one with flat road and another one with hilly road. We used the first track to validate our model. We came with new idea where we used edge detected images as the fourth layers for the train data set images(RGB and edge detected image). Please find the more details in report.
Udacity Simulator: https://github.com/udacity/self-driving-car-sim
drive.py: This file gets input frame data from simulator and feed it to trained model. Output of model is then sent to simulator for driving a car. This file is taken from https://github.com/ManajitPal/DeepLearningForSelfDrivingCars/blob/master/drive.py
We first generated training data from Udacity simulator and used it for training our CNN model.
References:
(1)https://towardsdatascience.com/deep-learning-for-self-driving-cars-7f198ef4cfa2 or https://arxiv.org/pdf/1604.07316.pdf (2)https://images.nvidia.com/content/tegra/automotive/images/2016/solutions/pdf/end-to-end-dl-using-px.pdf (3)https://github.com/udacity/self-driving-car-sim