In this repo, it contains the projects I worked on for the Self-Driving Car Nanodegree Course on Udacity. Each course is separated into directories. For each directory, a clone from the relevant repo is made, followed by the necessary edits required to constitute a complete submission for that project. You may consult each directory for the instructions for that particular project for more details.
moviepyopencv-pythontensorflow- Note that the course uses version 1.x but I will be using 2.x. Keras has been folded into Tensorflow from 2.0 and onwards so there's no need to install Keras separately. However, the exception to this rule is for the capstone project where it is required to use Tensorflow 1.x due to the requirements of the autonomous vehicle environment used for testing out the capstone.numpymatplotlibpandaspython-socketioeventletpillowflaskaiohttpjupyter
These can be installed through pip with the requirements.txt file included with this repo:
$ pip install -r requirements.txtFor the capstone project, please use the requirements.txt file located in the CarND-Capstone directory instead.
This project was cloned originally from: https://github.com/udacity/CarND-LaneLines-P1
This directory contains a method using simple computer vision algorithms to help localise where lanes are with respect to the ego vehicle. Given sample images from the front camera of a car on a highway, we determine the left and right lanes for the car. Please navigate to the CarND-LaneLines-P1 directory for more details. The report can be found in this directory under reflection_report.md.
This project was cloned originally from: https://github.com/udacity/CarND-Advanced-Lane-Lines
This directory contains a method using more advanced computer vision algorithms to help localise where lanes are with respect to the ego vehicle. The objective and criteria for performance evaluation are the same as Project #1. Please navigate to the CarND-Advanced-Lane-Lines-P2 directory for more details. The report can be found in this directory under reflection_report.md.
This project was cloned originally from: https://github.com/udacity/CarND-Traffic-Sign-Classifier-Project
This directory contains a method using deep learning to classify traffic sign images. We specifically explore the dataset used for training this system and provide an architecture that performs the classification well on the training, validation and test dataset. Please navigate to the CarND-Traffic-Sign-Classification-Project-P3 directory for more details. The report can be found in this directory under final_report.md.
Note: The dataset is not stored on this repo, but it can be downloaded yourself. Open up the notebook file in the CarND-Traffic-Sign-Classification-Project-P3 directory and run the cells that are relevant to downloading the data.
This project was cloned originally from: https://github.com/udacity/CarND-Behavioral-Cloning-P3
This directory contains a method using deep learning to output the steering angle for a car to ensure it stays on the road given a front-facing camera image. The data to collect was from a simulator provided by Udacity where we capture front-facing camera images as well as the steering angle so that we can train a deep neural network. This simulator is called the Term 1 simulator. This will be directly substituted into the simulator to see if we can autonomously keep the car on the road. Please navigate to the CarND-Behavioral-Cloning-P4 directory for more details. The report can be found in this directory under final_report.md. Once you get the simulator running (see the note below), all you have to do is run the Term 1 simulator, then execute the drive.py Python file with the supplied model in a separate terminal. Once executed, go back to the Term 1 simulator and choose track one and run this in autonomous mode to see the results.
Note: As of this writing, the provided simulator on https://github.com/udacity/self-driving-car-sim does not run on Mac OS Catalina. Fortunately, there were efforts made to rebuild the simulator on Catalina and can be retrieved here: https://github.com/endymioncheung/CarND-MacCatalinaSimulator. Just in case this repo no longer exists, I have forked it here: https://github.com/rayryeng/CarND-MacCatalinaSimulator.
Note #2: The model was trained using Google Colab Pro, so there is a training notebook you can examine to see how the model was trained. In addition, the dataset used for training this model is not available on this repo due to size constraints. It is hosted on my personal Dropbox so you can run the cell in the notebook to download it on your computer if you so desire.
Note #3: Be advised that Tensorflow 2.0 was used to train this model. The course uses TF 1.x with Keras 2.x using TF as the backend. TF and Keras were separated in older versions of the software. As of TF 2.0, Keras was eventually integrated into TF.
This project was cloned originally from: https://github.com/udacity/CarND-Extended-Kalman-Filter-Project
This directory contains an implementation of the Extended Kalman Filter (EKF) to track the position of a vehicle given LiDAR and Radar measurements. This is designed to work with the Term 2 simulator provided from Udacity. Please navigate to the CarND-Extended-Kalman-Filter-Project-P5 directory for more details. Once you get the simulator running (see the note below), all you have to do is run the Term 2 simulator, then execute the ExtendedKF executable when you build the project. Once executed, go back to the Term 2 simulator and choose Project 1/2: EKF and UKF, choose Select, then either Dataset 1 or 2 on the right side of the pane, then finally push the Start button to see the results.
Note: As of this writing, the provided simulator on https://github.com/udacity/self-driving-car-sim does not run on Mac OS Catalina. Fortunately, there were efforts made to rebuild the simulator on Catalina and can be retrieved here: https://github.com/endymioncheung/CarND-MacCatalinaSimulator. Just in case this repo no longer exists, I have forked it here: https://github.com/rayryeng/CarND-MacCatalinaSimulator.
This project was cloned originally from: https://github.com/udacity/CarND-Kidnapped-Vehicle-Project
This directory contains an implementation of the Particle Filter to determine the location of a car given landmarks and LIDAR measurements. This is designed to work with the Term 2 simulator provided by Udacity. Please navigate to the CarND-Kidnapped-Vehicle-Project-P6 directory for more details. Once you get the simulator running (see note below), all you have to do is run the Term 2 simulator, then execute the particle_filter executable when you build the project. Once executed, go back to the Term 2 simulator and choose Project 3: Kidnapped Vehicle, choose Select, then click on the Start button on the bottom right of the window.
Note: As of this writing, the provided simulator on https://github.com/udacity/self-driving-car-sim does not run on Mac OS Catalina. Fortunately, there were efforts made to rebuild the simulator on Catalina and can be retrieved here: https://github.com/endymioncheung/CarND-MacCatalinaSimulator. Just in case this repo no longer exists, I have forked it here: https://github.com/rayryeng/CarND-MacCatalinaSimulator.
This project was cloned originally from: https://github.com/udacity/CarND-Path-Planning-Project
This directory contains an implementation of a path planning algorithm where given an ego vehicle's position on the road, calculate a set of waypoints to allow the ego vehicle to change lanes, keep in their lane unless required to change lanes, follow the speed limit and not break acceleration and jerk guidelines. This is designed to work with the Term 3 simulator provided by Udacity. Please navigate to the CarND-Path-Planning-Project-P7 directory for more details. Once you get the simulator running (see note below), first run the the path_planning executable when you build the project. Once executed, run the Term3 simulator. Once executed, the path planning algorithm should immediately begin and you'll see the car start moving. Finally, the report that talks about the logic for the path planning can be found in final_report.md in Markdown and its PDF version in final_report.pdf in this directory.
Note: As of this writing, the provided simulator on https://github.com/udacity/self-driving-car-sim does not run on Mac OS Catalina. Fortunately, there were efforts made to rebuild the simulator on Catalina and can be retrieved here: https://github.com/endymioncheung/CarND-MacCatalinaSimulator. Just in case this repo no longer exists, I have forked it here: https://github.com/rayryeng/CarND-MacCatalinaSimulator.
This project was cloned originally from: https://github.com/udacity/CarND-PID-Control-Project
This directory contains a method for controlling the steering angle of the ego vehicle using PID controller. The goal is to provide the right steering angle so that the ego vehicle can autonomously go around a track once. This simulator is called the Term 2 simulator. Please navigate to the CarND-PID-Control-Project-P8 directory for more details. The report can be found in this directory under final_report.md. First build the project, then execute the pid executable. Once executed, run the Term 2 simulator, then execute the drive.py Python file with the supplied model in a separate terminal. Once executed, go back to the Term 2 simulator and choose Project 4: PID Controller, choose Select, then click on the Start button on the bottom right of the window.
Note: As of this writing, the provided simulator on https://github.com/udacity/self-driving-car-sim does not run on Mac OS Catalina. Fortunately, there were efforts made to rebuild the simulator on Catalina and can be retrieved here: https://github.com/endymioncheung/CarND-MacCatalinaSimulator. Just in case this repo no longer exists, I have forked it here: https://github.com/rayryeng/CarND-MacCatalinaSimulator.
This project was cloned originally from: https://github.com/udacity/CarND-Capstone
You can find the implementation of the final capstone project in CarND-Capstone. Please note that there is currently no way to run this on Mac OS Catalina. I had to resort to using the Udacity workspace to develop this project. Also note that the traffic sign detection algorithm was implemented in Tensorflow 1.3.0 because of the current hardware specifications of Carla, the autonomous vehicle that the project would be deployed on for testing. The trained models for the traffic sign detection have been committed via Git LFS so this will need to be installed if you want to grab the models.
Please visit https://github.com/git-lfs/git-lfs/wiki/Installation to install Git LFS. Once you do, after you clone this repo, please use the following to download the weights.
$ cd <path to this repo>
$ git lfs fetchThe weights will appear in CarND-Capstone/ros/src/tl_detector/light_classification/train_model. There are two files - one is the trained model for the simulator and one for actual live images where the model finally gets deployed on Carla.