Huawei’s genius teenagers create autonomous driving bikes! The drawing is open source, and the hardware cost is 10,000


Fish and sheep Jin Lei is from the Temple of Gafei Temple


Quantum Late Report | Public Account QBITAI

Sure enough, the master is in the folk.

Who can expect, a bicycle

Hard core

Open the way, it’s actually a fire



Kick your feet, the golden chicken is independent and stable as Taishan:

Add a brick with a unilateral side? Small scene:


After all, it is even difficult

“Acrobic Pose”

Bicycles that can be held at all:


Two steps for nothing?

Can there be, or you can still


The kind of:


On the road

, No pressure:

It is even equipped with Tesla

Laser radar


This is the “Wild Iron Man” station B


Zhihui Jun


time consuming

4 months

Time (basically just weekend), a new project created



In just a few days, the playback volume exceeds more than


, High hard core and complexity, look at the netizens directly.

Come and feel this style of painting.

Even the owner of the station B was attracted:

Even people in the industry called directly:

Any part of the part is an undergraduate thesis.

There are physical things, more than enough to graduate from a master’s degree …

How exactly does Zhihui Jun build such a “Cyberpunk” bicycle?

Next, let’s find out together (authorized).

(You may also care about its price, don’t worry, we announce the answer at the end of the article ~)


How many steps do you divide into autonomous driving?

The bicycle chosen by Zhihui Jun is very familiar with everyone

fixed gear

(Fixed Gear).

(Yes, it is the favorite of drift players)

A major feature of this kind of bicycle is simple structure and no brake.

So, Zhihui Jun first carried out a dead fly

CAD modeling



After the modeling work is completed, I got a

Virtual bicycle

All the later transformation work can be completed on the computer.

On the whole, the birth of Xuan is divided into a total


Three steps

Hardware reconstruction

: Because the bicycle is an arrears driver system, the first thing to do is to make it stabilize and then run. This step will involve a complex set of

Automatic Control System


: After all, you must be erased when you are autonomous driving. Therefore, this step is composed of a whole set of sensors

Perception network

And a computing chip with a strong computing power


Software and algorithm

: The last step is to develop and realize a set of perception and control algorithms on the basis of the above hardware. Zhihui Jun called it “injected soul”.

Step by step, first look at the first big problem.

How to maintain a balanced bicycle by yourself?

When riding a bicycle,


Front and rear

The balance is supported by the two wheels and ground friction.


Left -right direction

In the process of exercise, the left -right adjustment of the faucet is rely on the acceleration of the steering to offset the impact of gravity. like this:

What is in a state of still?

The idea given by Zhihui Jun is–


Simply put, when the car is dumped under the action of gravity, it uses a momentum wheel to generate a reverse torque to offset the gravity torque.

Specifically, Zhihui Jun installed a

Brushless Motor

The purpose is to drive a metal momentum.

In the back of the seat, he also installed a brushless motor. The purpose was to drive the rear wheel by friction and let the bicycle move the overall movement.

Zhihui Jun wanted to use 3D printing to complete the processing of these two parts, but because of the lack of material hardness and rigidity, metal materials became the first choice for this time.

(Fortunately, Zhihui Jun’s friend’s house “has a factory”, the problem is solved)

Then, there is a happy assembly link:

OK, the balance of bicycles is solved.

Bicycle’s “machine brain”


There is a hardware framework, and you want to let the bicycle run or even recognize the road. The next step is to “brain”.

Similar to humans, robots are also divided into “brain” and “cerebellum”.

Simply put, the cerebellum is used to control real -time action; the brain is responsible for consuming computing power, high delay perception and decision -making.

This time, Zhihui Jun’s “cerebellum” on the bicycle was ESP32.

ESP32 is a low-cost, low-power single-chip controller, integrated Wi-Fi and dual-mode Bluetooth.

Based on FRERTOS (an open source embedded real -time operating system), Zhihui Jun set up the RPC communication framework of the cerebellum to achieve the data processing of the sensor and the control algorithm of the motor.

The “brain” is ascending 310. This is the edge computing chip of the Shengteng series. Based on the Da Vinci architecture, it can provide a maximum computing power of 22TOPS.

The development of the brain is mainly implemented based on Shengteng AI stack and ROS (robotic operating system).

Therefore, with the help of deep cameras, lidar and other sensors, this bicycle can detect the surrounding objects, achieve obstacle avoidance and follow:


You can also identify the terrain witly and plan your own plan:

Bicycle’s “machine heart”

Of course, the bicycle of “Benefit” must not only have the mind, but also have a strong “heart” to drive it to work.

And Xuan’s “heart” comes from another project of Zhihui Jun,

CTRL-FOC vector control driver


This is a super mini dual -channel brushless motor drive, the maximum driver current can exceed



It is worth mentioning that this FOC driver of Zhihui Jun has two versions of the design scheme, one of which, the circuit design is the upper and lower folding board design, which separates the logical unit and the power unit. In other words, you only need to replace the drive circuit to match a larger power motor.


Specific to the project of Xuan, such a large driver in the figure below, finally achieved two.


The brushless motor FOC control also drove the front 60kg torque steering gear and heat dissipation fan.

Finally, inject the soul

Everything is ready, but I want to go on the road, so far I still have a little soul of alchemy–



Zhihui Jun said that there are more than 50 important parameters in the control code of this self -driving bicycle, including the control cycle, feedback matrix, PID gain, filter dead frequency, and so on.

In addition, physical parameters such as the power of the motor and the distribution of center of gravity will also affect the final effect.

With a little carelessness, the bicycle is not as stable as an old dog …

After the math modeling, Zhihui Jun also adopted the game engine Unity as the simulation environment. Like the research and development of serious autonomous driving, let the bicycle run down in the simulation environment.

In short, it is not easy to make a bicycle mature.

Some netizens said:

People are stupid. The topic of graduation design is the self -balance system design of the bicycle, so that it is not possible to adjust the ginseng.


However, due to the power of the vehicle, the bicycle is not yet available.

In addition, some netizens have proposed that the use of the rotation of the motion to control the balance belongs to a certain sense of cheating. In this regard, Zhihui Jun responded: The current design of the dragon -headed transmission structure is not reasonable, and then he will try to control the balance with only the faucet.

Why create xuan?


Kihui Jun, about “stack overflow engineers” and “wild Iron Man”, want to pay attention to the friends who want to pay attention to the technology zone of Station B. They are already familiar.

Zhihui Jun’s real name Peng Zhihui, graduated from the University of Electronic Science and Technology in 2018, joined Huawei through the “Genius Junior Plan” in 2020.

His videos have always been

It is known for example, from hand drawing PCB to writing C ++ code, all of which are available for small TVs.

This time, how did Zhihui Jun germinate the idea of ​​creating a self -driving bicycle?

There are some “violence” …

I accidentally fell on the bicycle, and my face brake on the ground …

Zhi Huijun said, “This cannot be just passed by”, so:

(A few seconds of distress Zhihui Jun)

Of course, Zhihui Jun himself also said in the video that in fact, he had the idea of ​​designing this bicycle very early.

At first, I saw the study of bicycle at Tsinghua University, which could follow humans and avoid obstacles agile.

The brain chip behind it


Also appeared on the cover of Nature in August 2019.

Curious friends will definitely ask,

What is the difference between the two



Quantum is done for you ~

first of all


Not the same.

Tsinghua’s bicycle is imitated with human behavior and does not use momentum control, while Zhihui Jun’s “Xuan” involves momentum control.

In addition, in terms of faucet, Zhihui Jun said that after improvement, Tsinghua’s control method can also be achieved (the dragon head balance is used in the movement process, and the static state uses the momentum balance).


But the core difference is that behind the bicycle created by Tsinghua, the core is

Brain chip

In this regard, Zhihui Jun said:

Their work is very powerful. I am incredible, but I have inspired them to make a similar application.

In order to reflect the difference from Tsinghua’s work, Zhihui Jun himself is still there

Project name

I ordered my efforts up and down.


“U” in xuan, which means


, “Evil translation” is -has not been nature …

This project is as open as ever, if you want to repeat …

Oh yes, it’s time to reveal Xuan’s



I -according to the incomplete statistics of the Great God himself,

Over 30,000 yuan

However, Zhihui Jun also said:

In the end, the hardware cost should be less than 10,000, and the laser radar is more expensive, mainly trial and error costs.

And, the focus,

Not sell

Ha ~ ~


Autonomous driving bicycle


Video connection:

Project connection:

Reference link:



– Finish –


Quantum QBITAI · Headline signing contract

Pay attention to us, and learn about the frontier technology dynamics for the first time


Hard core