Intro To Self
Driving Cars Nanodegree
Concept
I completed the Intro To Self Driving Cars Nanodegree by Udacity, my first foyer into the world of autonomous vehicles. I had the opportunity to sharpen my Python skills, apply C++, implement matrices and calculus in code, and touch on computer vision and machine learning all in the context of solving self-driving car problems.
Intro to Self-Driving Cars Nanodegree certification.
This page goes over the projects I completed and what I learned through this course.
Bayesian Thinking
Parallel parking a car in simulation.
Programming a histogram filter in Python.
The course began with a deep dive into Bayesian Thinking. Human drivers observe their environment when driving and make decisions based on the likelihood of events happening. Of course, a driver doesn't actually calculate the probability of a car passing behind another car, however we trust probabilities because it is the only way to take action in the midst of uncertainty. Autonomous systems such as self driving cars make similar decisions based on probabilities, however, instead of intuition and experience self driving cars calculate the probability of events using sensors and programming which reasons out what action to take. The two projects that I completed was the Joy Ride project where I had to program the simulation of a car in a drag race, going around a circular track and conducting parallel parking as well as the Programming A Histogram Filter In Python project where a Bayes filter that represents beliefs as a histogram was created in order to solve a localization problem. For instance, in the image above, you can see the belief a robot has of it's location converge to a single large circle at the same position as the red star (the ground truth of where the robot is located).
​
Kalman Filter
Running a Kalman Filter on LiDAR data.
Next, in Working with Matrices the focus was on two tools which are vital to self-driving car engineers: object oriented programming and linear algebra, understood through the completion of the Implement A Matrix Class and Kalman Filter projects. The Kalman Filter project is shown below.
Performance Programming In C++
The foundations of C++ were also introduced in this course. The goal of the Python to C++ project was translation: get a program written in Python, and translate it into C++. After this, I focused on Performance Programming in C++ where the course explored how to write good code that runs correctly and best, industry practices. Through the Optimizing The Histogram Filter project, I worked on making the python code I had translated into C++ run faster and more efficiently.
Navigation And Implementing A Route Planner
A-Star Route Planner
Navigating and Data Structures really tested and allowed me to work on algorithmic thinking by using data structures and algorithms. Through the Implementing A Route Planner project, I implemented an A-Star search algorithm to find the shortest path between a starting node and a destination node on a map.
Vehicle Motion And Control
Reconstructing trajectories using sensor data.
Next, the course took a look at Vehicle Motion And Control, where the use of calculus and trigonometry in self driving cars became apparent. Additionally, there was a lesson on how a self driving car uses various motion sensors to help it understand its own motion. At the end of this section in the Reconstructing Trajectories Using Sensor Data project, I used raw sensor data (which gave information about the distance driven, acceleration, and rotation rates of a car) in order to reconstruct a vehicle's trajectory through space. Using sensor data I was able to solve some odometry problems as shown in the notebook below.
Computer Vision And Machine Learning: Traffic Light Classifier
Identifying a yellow light using computer vision.
Finally, in the Computer Vision And Machine Learning section of the course, I learnt how a computer sees an image and how we can use machine learning to teach a computer to identify images programmatically. The final Traffic Light Classifier project tested my understanding and skill of these concepts by tasking me to programatically classify whether a traffic light is red, yellow or green with an incredibly high accuracy of 95-100% (as a driver's life is dependent on such a classification) using computer vision and machine learning.