Written by Zheng Wen
Edited by Zhou Changxian
In any given industry, if a company is able to capture 80% market share, it’s likely an outlier in terms of its organizational structure.
DJI is a perfect example of this phenomenon.
With its superior drone products, DJI has become one of the few Chinese companies to cement its reputation in the European and American markets. In the popular US TV show “The Big Bang Theory”, Sheldon and Howard play with DJI’s Phantom drone for an entire episode.
During the hiring process at DJI, there is a “lethal character test” that some candidates may fail multiple times. They might wonder, what exactly does DJI look for in job candidates? These tests have no right or wrong answers, they simply help weed out candidates who do not fit the company’s culture.
For example, if you were in a meeting and your boss misunderstood a concept, what would you do?
a. Directly point out your boss’s mistake to speed up progress and avoid wasting time.
b. Wait until after the meeting is over to privately explain the error to your boss.
c. Do nothing.
Individuals with different personalities may have significantly different approaches to this scenario, and one job candidate chose option a.
Until now, it has been challenging for new hires to develop a reliable approach to passing the interview process. However, employees who have been with the company for some time tend to agree that their colleagues and managers alike are motivated and goal-oriented. One of DJI’s suppliers says that the company’s workforce is “paranoid” and “practical.”
Engineers may refer to themselves as geniuses, and they may refer to their product managers as gods. They have a nearly religious belief in their products and a drive that transcends mere monetary gain.
So, it’s no surprise that the market has recognized DJI’s brilliant engineers. In 2012, DJI introduced the world’s first aerial photography drone – the Phantom – which quickly became popular in the market. Subsequently, each new product released receives increasing recognition within the market.
However, the competition in the drone market increased rapidly, leading DJI founder Frank Wang to predict in 2016 in an interview that the drone market will soon reach saturation and that the company’s revenue will hit 20 billion yuan. As predicted, the drone market slowed down in 2019 as it approached its limits.
“We have done everything we can. We’re approaching the limit of the industry”, said a former DJI R&D employee.
Or have they?
Recently, DJI has given a new answer with its involvement in the “smart driving” industry, providing new solutions for the Kiwi EV 2023 model produced by SAIC-GM-Wuling. The engineers seem to be dipping their toes into a new Eden.When communicating with the media, the collaborating party stated their ambitions with just one sentence. “Both SAIC-GM-Wuling (SGMW) and DJI are aiming to bring smart driving technology into ordinary households,” said He Yibo, Vice General Manager of the SGMW Technology Center.
One sentence, “bringing smart driving technology into ordinary households,” is both literary and stern.
“Both parties believe that practical creators will popularize smart driving technology and eventually make it a standard feature in all passenger cars. Smart driving technology is not unattainable, but it requires extra cost and effort to experience its functions,” emphasized Xie Kandi, Director of PR at DJI Car.
Indeed, very practical.
Last year, the industry was still arguing about whether it was possible to produce a smart car for less than RMB 150,000. This year, kiwi EV, priced at around RMB 80,000, provided an answer through its actions.
From Drones to Intelligent Driving Systems
DJI, which was founded in Shenzhen in 2006, started as a research institute investigating robots and automation technology.
Within DJI, engineers often describe the company’s atmosphere as being like a laboratory. “It’s like doing scientific research. Even if you fail, you can start over again. No one is urging you to deliver quickly.”
DJI’s drones sprang from a line of research on robots and automation technology.
At first, they worked on the fly control. Based on interest, they hoped that an aircraft and a model aircraft could hover stably. To achieve stable hovering, one needs to have a range of software, hardware, electronic governors, controllers, IMUs, and fusion algorithm capabilities.
That’s a simple goal.
In late 2010, DJI had all the technology needed for an integrated drone capable of stable hovering, autonomously controlling its posture, and executing filming missions. However, it wasn’t until 2012 that the mature product was realized with the release of the Phantom drone, which became an instant hit.
Car technology and drones came from the same source. The self-driving force that propelled the development of car technology is similar to that of drones.
By 2016, the binocular vision and visual fusion positioning system, combined with some intelligent obstacle avoidance systems, were already being mass-produced for use in drones. At that time, the related car technology was still in the research and development stage, but DJI had already preliminarily possessed the technological capabilities, and the relevant technologies were indeed applied to the intelligent driving system.Xie Tiane reveals that DJI started talking to automakers about cooperation in 2018 and officially launched DJI Automotive BU in 2019. However, it wasn’t until the Shanghai Auto Show in April 2021 that it was officially announced to the public. The reason is simple: by then, DJI had already achieved mass production capability.
DJI’s “Lingxi Intelligent Driving System” for cars is an assisted driving system that focuses on solving the pain points and high-frequency scenarios of urban driving and parking, with core advantages in urban driving scenes and parking.
Two notable capabilities are:
- Identification of obstacles of any category
Thanks to DJI Automotive’s binocular vision system, it can recognize obstacles of any category without the need for prior learning or training. Based on algorithms utilizing binocular fusion point clouds, the system obtains geometric information including depth information to determine if an obstacle in front poses a threat to driving safety and responds accordingly, without understanding what the obstacle actually is.
- Support for parking in any angle and direction
Most vehicles with automatic parking on the market now require driving to a specific position and angle relative to the parking space to complete subsequent automatic parking. The Lingxi Intelligent Driving System, on the other hand, can automatically park in any space that it detects and complete the parking quickly.
Another point worth mentioning is that, in terms of safety, the Lingxi Intelligent Driving System has designed a system that gradually downgrades functions for consumers. In addition to ensuring safe completion of driving scenarios, in scenarios where the system is not developed or truly unable to respond, the system will engage its preset downgrade strategy and actively remind users to take control to ensure transfer of driving control is completed safely.
Specifically, the Lingxi Intelligent Driving system’s downgrade strategy is divided into three levels: the first is the lightest level, with an instrument display and text prompts; the second level involves interactive warnings and disabling some functions; and the most severe level, the third level, will automatically slow down and stop the vehicle gradually and activate double flashers and the handbrake if the user does not respond, ensuring driving safety.
“Basic Skills” Are Important
So, what are the essential conditions for achieving higher-level intelligent driving in the future? In the DJI team’s vision, basic skills and an integrated architecture are key.
First and foremost, basic skills are essential.
This system must be able to complete the core capabilities of basic ADAS without relying on high-precision maps or expensive sensors, which is an essential requirement.The Lingxi intelligent driving system, which uses binocular technology, does not rely on external data sources such as lidar or high-precision maps.
“Many city roads do not yet have high-precision maps. To ensure that this system is available in all scenarios, we need the system itself to have online sensing,” said Xie Qiandi. High-precision maps and lidar are shortcuts, and the basic perception and processing capabilities of the system are not enough, so they can first equip the system with hardware and data to achieve the required functionality.
Sensors, algorithms, computing power, and data are the four major challenges of autonomous driving. It is impossible to achieve the intelligent driving era by focusing on a single aspect. Instead, it requires joint optimization and progress in all aspects.
According to DJI, the so-called lidar camp and binocular vision camp are not actually contradictory in engineering practice. Looking ahead, the key behind the progress is basic skills, or the ability to achieve the core functionality with the most basic hardware.
In other words, DJI is not pursuing the best autonomous driving system with unlimited costs, covering as many corner cases as possible. Instead, it aims to polish the main scenarios with limited resources, based on engineering and mass production costs.
Xie Qiandi described it to AutocarMax: “Hardware, external data sources, and additional configurations are like magic weapons. If you don’t have enough internal strength, you can complete many complicated challenges with magic weapons, but they are very expensive and have many limitations. A genuine master can use fallen leaves and flying flowers in many situations, even without magic weapons. If you have the added support of magic weapons, the limit will be very high.”
Second, an integrated architecture is required.
Some traditional models have installed a separate set of sensors and controllers for each function, which cannot connect the functions together or continue to upgrade them, making it impossible to perform OTA upgrades.
Xie Qiandi emphasized that there is still great potential for improvement in the Lingxi intelligent driving system in this vehicle.
So, how is DJI’s smart car different from its competitors? He summarized two points:
First, they have accumulated years of experience and production capacity in designing and developing intelligent systems, with a production scale that can reach millions.
Second, they have a highly vertically integrated supply chain for both software and hardware.
Xie Qiandi believes that “practice and validation of engineering capabilities are the biggest challenges.”
For example, the binocular solution has a rigid connection, and the position, angle, and consistency must be highly consistent during mass production, requiring strict requirements for factory calibration and high-precision production.”DJIA has a shipment of millions, years of experience in intelligent systems, from design, development, production, this is the experience of engineering capabilities, with a car-level factory, able to fully implement their designs, this is our advantage,” said Xie Kandi.
“It can be said that the advantages of DJIA’s onboard system are aimed at solving problems, and we hope to solve the fundamental problems of the upcoming era of intelligent driving.”
As for how much imagination space there is for DJIA’s “smart driving”?
Xie Kandi made a metaphor, DJIA’s car system is an experienced chef, who can make a homely stir fry, and can also make a Chinese banquet. “You can’t say that because you made a homely stir fry, you didn’t bring up the elements of the Chinese banquet in the stir fry.”
“If in the future we need to make a product with a higher price positioning, higher user requirements, higher experience and special scenario capabilities for the intelligent driving system, we will define all the products of the Chinese banquet, and it is still an engineering problem.”
Perhaps, as DJIA predicted, the road to the popularization of intelligent driving systems will be more solid.
Of course, this depends on whether consumers are willing to pay for it, just like drones. At present, DJIA has high expectations for the Kivi EV 2023 model.
This article is a translation by ChatGPT of a Chinese report from 42HOW. If you have any questions about it, please email bd@42how.com.