Self-driving-car

The project demonstrates how to train a virtual car to drive autonomously in Unity simulator using reinforcement learning and random forest model.

Installation

Install Python, Unity, and Anaconda. Install the required libraries.

drive.py – This file implements socket communication and it is used to connect with Unity Environment. In autonomous mode the car predicts next action using q_learning_model.h5 and pos_detect_model.h5 file.
check_and_train.py – Code written in this file saves recorded images into an info.CSV file and trains pos_detect_model and q_learning_model. Trained models are saved into .h5 file.
action.py- This class defines several actions and calculate their values.
q_learning_model.py – In this file neural network is built and trained using reinforcement learning. States and actions of recorded images are stored using ReplayMemory class.
data_utils.py – This file contains utility methods such as loading data, image processing etc.
pos_detect_model.py – In this file neural network is built and trained using recorded images to predict the position of a car.

Install python, unity, and download Git lfs for the udacity environment.
After installation, you need to enable it on your machine: git lfs install
Next, download developers code from GitHub. https://github.com/suoeryu/CPSC587Project
Then create a python environment used by the model using conda: cd car-behavioral-cloning conda env create -f environments.xml
Activate car-behavioral-cloning using following commands: a. $ activate car-behavioral-cloning
Start the simulator, and a startup screen pops up.

To run the car in autonomous mode, run following command drive.py -m drive

To record and train model, follow below steps

Create a folder for storing images.
Go to the terminal and run following command to record images in training mode. python drive.py -s “”
Update the image folder path in check_and_train.py line 24.
Run check_and_train.py to train the models.
Run the car in an autonomous mode using following command drive.py -m drive

The code is derived from car-behavioral-cloning. Thank a lot Naoki Shibuya!