Four years ago, when I brought my little robot to Suzhou for a maker competition, I met Momo and Ni. Four years later, as they start their businesses and I am working in California, doing research, I hope to share my insights in AI with those who follow Darwen.
> Technology enthusiast, innovation lover, enjoys playing guitar and harmonica
> Graduated from UESTC, currently working at OPPO Shanghai Research Institute AI Laboratory as an AI algorithm engineer
> Currently focused on self-developed deep learning inference frameworks and heterogeneous acceleration for mobile computing platforms
> Rich experience in robot hardware and software, as well as machine learning technologies
> Hobbyist in industrial design, with a self-proclaimed aesthetic sense, knows only how to use Rhino for modeling
> Moderator of the Arduino Chinese community, enjoys crispy noodles and cola, and has a firm stance against eating cauliflower
> Personal website: http://www.pengzhihui.xyz/
I am Zhihui, officially joining “Darwen” today, and will share cutting-edge knowledge in artificial intelligence from time to time.
To briefly answer today’s question, here are some of my suggestions:
-
Start with Arduino, find a few open-source projects to familiarize yourself with the development process, and most importantly, cultivate interest and a sense of achievement.
-
What hardware platform and tools to use depend on what you want to create; there are no universal tools.
-
Do not limit your knowledge boundaries; no one says you cannot use Raspberry Pi while working with microcontrollers. The world of electronics is full of joy, and every inch gained brings happiness.
This is my first time meeting everyone; let me share how I embarked on my journey in AI.I started learning about microcontrollers in my sophomore year, and I have been working with hardware for about five or six years now. Around my senior year, I became interested in Linux and operating systems, and switched my graduate studies from EE to CS. Currently, I am doing research on deep learning algorithms at a certain green company’s AI laboratory. So whether it’s Arduino or Raspberry Pi, I have explored both thoroughly.
I am a mini-robot enthusiast and also an electronics major. I enjoy creating various robot projects and have open-sourced some of my work.
I remember getting a Vector robot shortly after it was released; it is currently one of the smartest toy robots available—cute in appearance, with a rich array of sensors (camera, microphone array, touch sensors, infrared sensors, IMU), and the official SDK is generously open, allowing for programming with Python to implement some features, which is quite playable. However, after playing with it for a week, I sold it because I couldn’t modify the hardware according to my ideas, which was unbearable!
Thus, my first mini self-balancing robot, Egg Yolk 1.0, was born, with a very simple function: self-balancing (the principle is similar to that of the Xiaomi balance bike). This little robot opened the switch to my subsequent journey in electronic design and led me to learn Arduino and self-study STM32, writing software libraries and soldering various component packages on circuit boards. It has been five years since then.
👆This little robot opened the switch to my journey in electronic design. After learning Arduino and self-studying STM32, writing software libraries and soldering various component packages on circuit boards, and mastering Android APP development and 3D modeling skills, I could no longer be satisfied with such crude designs. Two years later, the Egg Yolk welcomed the 2.0 version updateNano, which turned out like this👇
The production of this robot can be said to have utilized all the knowledge I had at that time. I have a peculiar obsession with the compact nature of robots, and achieving such integration without designing my own PCB and merely using ready-made modules truly moved me…
With the previous accumulation, Egg Yolk 3.0 came along, a small robot similar to the Vector, created without any open-source ARM board, but directly based on an idle Android phone (thanks to the rapid development of mobile internet, most phones today perform far better than Raspberry Pi), achieving voice dialogue and visual functions. This was what I was working on three years ago.👇
Although the Egg Yolk 1.0, 2.0, and 3.0 look complex, they are ultimately the products of 8-bit microcontrollers (also known as MCU), and the term “intelligent” carries too much weight in the world of microcontrollers. Resources on Arduino, 51, MSP430, and even the ARM family’s STM32 are too limited, with only deeply trimmed and optimized RTOS capable of running, which cannot meet the needs of complex programming.
All versions of the Egg Yolk series robots are open-sourced on my personal website: http://www.pengzhihui.xyz/
So where is the Egg Yolk 4.0 version? It is definitely striving for higher integration and more intelligence. Is it reasonable to use Raspberry Pi for the Egg Yolk 4.0 version? Although Raspberry Pi’s performance is not on par with high-end SoCs like Qualcomm or MTK, it at least runs Linux, which supports open-source frameworks like OpenCV and TensorFlow for computer vision and machine learning, making it very reasonable to develop intelligent applications.
I have graduated now, and my spare time is not as plentiful as when I was in school. Although my current job involves machine learning algorithms, seemingly unrelated to hardware, my passion for robotics has not diminished at all. Being able to create a robot from a pile of resistors, capacitors, and messy code always gives me a sense of joy akin to creating life.
If you are interested in robotics, you must have the spirit of “being willing to tinker.” A few days ago, I watched Boston Dynamics’ Atlas evolve from a young toddler to now being able to “run up walls,” and I genuinely felt my blood boil.
For myself, beyond the body of the robot, what I want to do more is to make it smarter and have an “interesting soul,” which is also one of the reasons I entered the AI industry. The future of the AI world belongs to cloud and edge computing, as industry giants are launching their edge computing hardware, among which Raspberry Pi is well-known.
Ni Mo: Besides that, Intel’s Movidius is a great platform for intelligent development. If many hardware engineers have a strong passion for embedded hardware and AI algorithms, we would like to introduce the Movidius™ Myriad™ X VPU based NCS2.The future Egg Yolk 4.0 version will likely use NCS2 for deep computing, and we are just as excited as everyone else.
