Where To Go After Andrew Ng Deep Learning

DeepPiCar Series

DeepPiCar — Part one: How to Build a Deep Learning, Cocky Driving Robotic Motorcar on a Shoestring Budget

An overview of how to build a Raspberry Pi and TensorFlow powered, self-driving robotic car

Introduction

Today, Tesla, Google, Uber, and GM are all trying to create their ain self-driving cars that can run on real-world roads. Many analysts predict that within the next v years, we will start to have fully autonomous cars running in our cities, and inside thirty years, nearly ALL cars will exist fully autonomous. Wouldn't it exist cool to build your very ain cocky-driving car using some of the same techniques the big guys use? In this and next few articles, I will guide you lot through how to build your ain concrete, deep-learning, self-driving robotic auto from scratch. You will be able to make your machine detect and follow lanes, recognize and respond to traffic signs and people on the route in under a week. Hither is a sneak peek at your final product.

Lane Post-obit (left) and Traffic Sign and People Detection (right) from DeepPiCar's DashCam

Our Route Map

Part ii: I will listing what hardware to buy and how to set them upward. In short, yous volition demand a Raspberry Pi board($50), SunFounder PiCar kit ($115), Google's Edge TPU ($75) plus a few accessories, and how each part is important in later on manufactures. The full cost of the materials is effectually $250–300. We will likewise install all the software drivers needed past Raspberry Pi and PiCar.

Raspberry Pi 3 B+ (left), SunFounder PiCar-V (middle), Google Edge TPU (correct)

Part 3: We volition gear up up all the Estimator Vision and Deep Learning software needed. The main software tools we use are Python (the de-facto programming language for Automobile Learning/AI tasks), OpenCV (a powerful computer vision package) and Tensorflow (Google'southward pop deep learning framework). Note all the software we utilise hither are FREE and open source!

Part four: With the (tedious) hardware and software setup out of the way, nosotros will dive right into the FUN parts! Our get-go project is to use python and OpenCV to teach DeepPiCar to navigate autonomously on a winding single lane road past detecting lane lines and steer accordingly.

Part 5: we volition railroad train DeepPiCar to navigate the lane autonomously without having to explicitly write logic to control information technology, as was done in our first projection. This is accomplished by using "beliefs cloning", where we use but the videos of the route and the correct steering angles for each video frame to train DeepPiCar to drive itself. The implementation is inspired by NVIDIA's DAVE-2 full-sized autonomous motorcar, which uses a deep Convolutional Neural Network to notice road features and brand the right steering decisions.

Lane Following in Action

Lastly, in Office half dozen: Nosotros will apply deep learning techniques such equally single shot multi-box object detection and transfer learning to teach DeepPiCar to discover various (miniature) traffic signs and pedestrians on the road. And and so we will teach it to end at scarlet lights and cease signs, become on light-green lights, cease to wait for a pedestrian to cantankerous, and alter its speed limit according to the posted speed signs, etc.

Traffic Signs and People Detection Model Grooming in TensorFlow

Prerequisite

Here are the prerequisites of these articles:

First and foremost is the willingness to tinker and break things. Unlike in a car simulator, where everything is deterministic and perfectly repeatable, existent-world model cars can be unpredictable and you lot must be willing to get your hands dirty and start to tinker with both the hardware and software.
Basic Python programming skills. I will presume you know how to read python code and write functions, if statements and loops in python. Most of my code is well documented, specifically the harder to understand parts.
Basic Linux operating system cognition. I will assume y'all know how to run commands in Bash shell in Linux, which is Raspberry Pi's operating system. My articles will tell you exactly which commands to run, why we run them, and what to look as output.
Lastly, yous will need about $250-$300 to buy all the hardware and working PC (Windows/Mac or Linux). Again, all the software used volition be gratis.

Further Thoughts [Optional]

This is optional reading, as I try to cover everything you need to know in my articles. However, if you want to dive deeper into deep learning, (pun intended), in additional to the links I provided throughout the article, hither are some more resources to bank check out.

Andrew Ng'southward Machine Learning and Deep Learning courses on Coursera. Information technology was these courses that ignited my passion for Machine Learning and AI, and gave me the inspiration to create DeepPiCar.

Automobile Learning (FREE): This grade covers traditional Motorcar learning techniques, such as Linear regression, Logistic regression, and Support Vector Machines, etc, besides equally Neural Networks. It was created back in 2012, so some of the tools information technology uses, namely Matlab/Octave, are out of way, and it didn't talk about deep learning in keen length. But the concepts it teaches you lot are invaluable. Yous merely demand loftier school level math and some basic programming skills to take the course and Dr. Ng explains hard concepts similar backpropagation extremely well. It takes about three months to complete this course.

Deep Learning 5-Course Specialization (FREE or $50/calendar month if you want to get the document): This course was introduced in early on 2018. And so it covers all the latest AI enquiry up to that time, such as Fully Connected Neural Networks, Convolutional Neural Network (CNN), and Sequence Models (RNN/LSTM). This course was such a treat for me. Equally an engineer, I always wonder how some of the cool gadgets piece of work, such every bit how does Siri respond to your questions, and how does a motorcar recognize objects on the road, etc. Now I know. It takes about 3–4 months to complete this 5-course specialization.