Car stuff (public number: chedongxi)
Author | Xiaohan
Edited by | Xiaohan
Another self-driving company gets financing!
A few days ago, L4 Robotaxi company Lingjun Technology announced that it has obtained a new round of financing of tens of millions of yuan, and Tianyancha industrial and commercial information shows that the investment is led by the freight industry "Didi" - manbang Group, a LISTED company in the United States, and followed by the old shareholder Zhenxin Capital.
Although Lingjun Technology was not well-known before, it can be seen from the recent automatic driving financing news, including this time, that all kinds of investors are increasingly favoring potential companies with latecomer advantages.
This phenomenon is not difficult to understand. After several years of development, the head players face three major dilemmas after reaching the valuation high point: the US stock listing path is not smooth; the technology has not opened a significant gap with the players in the rear, and the realization of unmanned driving is still far away; the lack of hematopoietic ability, the company is always facing the risk of food shortage...
In short, the industry has entered a long and painful marathon stage, and investors from all walks of life have naturally become more and more interested in those who seem to be running slower, but have good team and technical capabilities.
In the words of Yang Wenli, founder and CEO of Lingjun Technology, "Others spend billions to achieve the effect (Robotaxi), we spend tens of millions to do it." ”
▲Lingjun Technology CEO Yang Wenli
Although the words are not long, the words and lines all reveal his confidence in his own technology, and at the same time reflect that Lingjun Technology has a different understanding and playing style of autonomous driving entrepreneurship. After several hours of exchanges with Yang Wenli and Pang Dongjun, COO of Lingjun Technology, Che Dong was able to show you more information about the old entrepreneurial team of Lingjun Technology and deeply interpret this "slow" development model.
First, the competition for autonomous driving is after 2025
Lingjun Technology was established in 2016. In terms of time, it is completely among the first startups in the industry, but in the past 6 years, it has been unusually low-key, even a little marginal.
"On the surface, we have a small team with less money, but in fact, we deliberately slow down the speed of development." Referring to the development issue, Yang Wenli explained to Che Dongxi, "We are actively controlling the speed of development. ”
▲Lingjun Technology's autonomous driving test vehicle
This is directly related to Yang Wenli and the founding team's judgment on the development of the industry. In their view, the development of the autonomous driving industry is divided into four periods:
1. The maturity period of the core supply chain
The time frame is 2013-2020. At this stage, the core supply chain of autonomous driving is not yet mature and expensive, and the main job of each company is to polish the technology and try to mass-produce it.
2. Small-scale commercialization landing period
The time frame is 2021-2025. At this stage, with the increase of demonstration cities, technical products such as unmanned minibuses, intelligent network buses, and regional unmanned logistics vehicles can be applied on a small scale, and enterprises can obtain certain income.
3. The period of market scale expansion
The time frame is 2026-2030. At this stage, as the technology becomes more mature, the regulatory regulations are further opened, the autonomous driving technology begins to land in more and more scenarios and scopes, the fierce competition between the companies, and the industry pattern begins to reshape.
4. Period of international expansion
The time frame is after 2030. At this stage, as domestic competition is approaching the end, the pattern is basically stable. Players need to open up the international market to further develop, and at the same time need to refine the operation of a large technology company.
Such industry cognition guides the development ideas and rhythm of Lingjun Technology.
For example, the first stage of the industry is generally facing the problem of lack of wire control chassis and lidar is not easy to make. At this time, it is obviously not suitable for large-scale expansion, if a lot of people are recruited to spend a lot of effort to solve these parts problems, in the second stage, it will be found that the supplier has solved the above problems, which is equivalent to the useless work done before.
Another example is the second stage, which is not suitable for large-scale expansion when the technology is not mature enough.
"At this stage, if hundreds or even thousands of test vehicles are running all over the country, once the technical structure changes or major upgrades, the existing fleet will become an invalid asset and it will be difficult to deal with." Yang Wenli said.
There are obvious cases of this phenomenon at home and abroad. For example, Waymo released the fifth generation of Robotaxi this year, which has been upgraded in the vehicle body and sensors and computing platforms, while the old model cannot be directly upgraded to the fifth generation, and the fleet needs to be gradually replaced.
In addition, the autonomous driving industry has just entered the second stage, and the overall revenue capacity is not strong, and it has to rely on financing and blood transfusion.
"Large and high valuations will lead to increasingly difficult financing in the future, and finally it will have to lay off employees or go bankrupt." Yang Wenli analyzed the car and said, "But the small and beautiful team can achieve breakeven in the second stage and ensure that they live to the next stage." ”
Second, start with "Baidu autonomous driving 20 people"
One day in 2015, a BMW 3 Series GT painted in orange and white with a "big flower pot" on the roof drove onto the Beijing G7 Highway, then turned into the fifth ring road, turned around and returned to the Headquarters of Xierqi Baidu.
This BMW car is the starting point for domestic commercial autonomous driving research and development. It was built by a team of more than 20 people in Baidu and a BMW team. In the past five or six years, these more than 20 people have left Baidu one after another and created a number of autonomous driving companies, supporting half of the domestic autonomous driving industry.
Yang Wenli is one of these 20 people, and he is an uncompromising autopilot veteran.
He completed his undergraduate and master's studies in the Department of Automation at Tsinghua University, then received his Ph.D. from Pennsylvania State University in the United States, and then worked at Western Digital as a chief architect.
After returning to China, Yang Wenli joined Baidu's cutting-edge deep learning research institute at that time, and participated in the creation of Baidu's early autonomous driving team.
In 2016, Yang Wenli and his Tsinghua alumni Yan Han and He Jiarui founded Lingjun Technology, and the research and development goal was locked in the most difficult L4 grade Robotaxi direction.
Yan Han holds a Ph.D. degree from Tsinghua University Department of Automation and is a senior vice president of map and simulation research and development at Lingjun Technology. It is a continuous successful entrepreneur, the first startup company involved in after graduation has been listed, and the company that joined Lingjun Technology for the second time was acquired by Qihoo 360 and served as a senior research and development engineer of AI, responsible for the basic technology research and development of operating system simulators, network security, pattern recognition and other fields.
He Jiarui is a master's student in the Department of Automation at Tsinghua University, previously a senior R&D engineer in Baidu's Autonomous Driving Division, and currently served as the vice president of R&D of senior decision planning and development of Lingjun Technology.
He holds a ph.D. from the University of Tokyo, and is the former founding partner of Linglu Technology and CTO Si Ruochen, who serves as the senior vice president of perception system research and development of Lingjun Technology.
In addition, Lingjun Technology also has two partners, Deng Haiqing and Pang Dongjun, who play the roles of CFO and COO respectively. Deng Haiqing is a postdoctoral fellow of the People's Bank of China, an independent director of the Bank of Ganzhou, and a visiting professor at many universities such as the National People's Congress and Beijing Normal University, and has rich financial experience.
▲Pang Dongjun, COO of Lingjun Technology
Pang Dongjun has worked in the autonomous driving industry for many years. He has successively been responsible for business and key customer business in Tage Zhixing and Yikong Intelligent Driving, and then co-founded in Wanzhi Driving as vice president of commerce, with rich experience in technology application landing.
From the background of the core members, it can be seen that while Lingjun Technology has strong technical capabilities, it also focuses on introducing professional talents in finance and technology landing, and builds a perfect team with complementary advantages.
The high-quality team has naturally been continuously sought after by the capital. In 2017 and 2021, it successively received angel round investments from Wuyuefeng, Jiuhe and Xintian and Pre-A round financing from Zhenxin Capital. In addition, well-known domestic AI chip companies such as Horizon and Ganzhou Development and Investment are also its investors.
In addition, it should be noted that the 6 years since the establishment of Lingjun Technology are the most lively time in the global autonomous driving industry - the wave of entrepreneurship has risen and fallen, and large amounts of financing have continued. Over the past 2,000 days, various peers, car companies and mobility companies have been throwing olive branches to Lingjun Technology or its team members in the hope of acquiring or poaching its technical experts.
However, its core team is completely unmoved by short-term interests, firm in its development direction, and no one has left so far, behind which is a clear industry judgment and tenacious strategic determination in play.
Third, Robotaxi to get the urban scene Robobus normal operation
In terms of technology research and development, the development strategy of Lingjun Technology is to "extract pearls and reduce dimensions for commercial use".
In the view of Yang Wenli and the Lingjun technology team, the unmanned (that is, Robotaxi) technology in the urban scene is the pearl in the crown of the industry, so the company has locked in this direction from day one, hoping to finally pick the pearl.
However, because Robotaxi is the most difficult technology and it will take a long time to achieve, it is also necessary to apply the technology dimension reduction in specific scenarios in the research and development process. One is to earn money to feed the team through commercialization, and the other is to collect data to promote the iteration of Robotaxi technology.
Adhering to this development idea, Lingjun Technology has selected two major scenarios of Robobus autonomous bus and urban regional logistics for dimensionality reduction application.
The bus and regional logistics vehicle driving scene is exactly the same as that of Robotaxi, but the driving speed of these two is relatively slow, and most of them are fixed routes, so it is very suitable for creating products based on Robotaxi technology and landing commercially.
Pang Dongjun, COO of Lingjun Technology, told Chedong that it has built two Robobos, namely unmanned minibuses for the park and intelligent network buses with a length of 5.9 meters, and both have opened normal operations in the new energy vehicle science and technology city in Jiangxi Ganzhou Economic Development Zone.
▲Lingjun Technology's self-driving minibus
Unmanned minibuses are low-speed vehicles, there is no steering wheel and operating pedals in the car, and they mainly operate in semi-enclosed scenes. Intelligent networked buses follow a fixed route between airports, railway stations and economic development zone management committees.
During its operation, its uninhabited minibuses had up to 3 day shifts without any takeovers. Its working hours per day shift are 8 hours, and the operating mileage is about 150 kilometers, in other words, it has traveled 450 kilometers in 24 hours in the urban scene without any takeover, which shows that its automatic driving technology has reached a high level of maturity.
It is worth mentioning that Lingjun Technology adopts a group development strategy. In the landing area, in addition to unmanned minibuses and intelligent networked buses, its Robotaxi will also be driven to test. All three models are tested simultaneously to collect data, which in turn facilitates technology iteration, which in turn improves the driving performance of all three models at the same time.
▲Lingjun Technology autonomous driving minibus
Since November last year, the three test vehicles have been in regular test operation for 5 months.
The performance of the Robotaxi model of Lingjun Technology is also remarkable. According to Yang Wenli, although the scale of its Robotaxi fleet is small, it already has the ability to achieve P2P (from parking lot to parking lot) in the city, which can handle all driving scenarios including parking, ordinary roads, intersections, highways, roundabouts, tunnels and so on.
Fourth, with the end as the beginning of a variety of innovative ways to develop unmanned vehicles
"Others spent billions to achieve the effect, we spent tens of millions to achieve it." When talking about the technical performance of his own unmanned vehicle, Yang Wenli smiled and gave such an evaluation to the car.
The biggest difficulty of Robotaxi, or L4 autonomous driving in the city, is mainly the game problem with other traffic participants, that is, the scale of decision-making. Lingjun Technology responds to this problem through a number of technological innovations such as hybrid decision model, data rumination, simulation testing, etc.
In urban scenarios, unmanned vehicles will encounter almost endless special situations, and rule-based decision-making models simply cannot cope with them.
Lingjun Technology then integrates rules and AI technology, which learns how human drivers deal with various special situations through deep learning technology, and then uses rules (such as not compacting lines or running red lights) to ensure that the driving decisions given by AI algorithms are safe and in line with the rules.
In this way, it not only gives full play to the advantages of AI technology to solve special problems, but also avoids the black box problem of deep learning models, the best of both worlds.
Of course, the premise of the above approach is to have enough special scenario data to facilitate technical iteration. The size of lingjun technology's fleet is not much, where does the data come from? What happens to the data? And how to apply data for iteration?
Yang Wenli introduced a data research and development closed loop.
First of all, products such as Robobus, which operate normally, collect a lot of data every day. After obtaining the data, Lingjun Technology will conduct semantic-level analysis of the data, extract multiple independent scenes, and build an incremental scene library.
Secondly, it will put the scene library into the simulation engine and randomly arrange and combine with the existing scenes to reconstruct hundreds of kilometers, or even thousands of kilometers of virtual test scenes - greatly improving the efficiency of data application.
Finally, with a larger-scale test scenario, Lingjun Technology can use this to polish its own algorithm, improve the performance of autonomous driving products, and form a data research and development closed loop of "product-data-scene-simulation-algorithm-product".
In addition to the closed loop, Lingjun Technology's research and development ideas that start from the end are also quite bright.
In Yang Wenli's view, the ultimate and most important goal of the automatic driving system is to make good decision planning, so the essence of building a good automatic driving system is to create a good decision planning system.
In addition, R&D technology must take into account the mass production of software and hardware from the beginning - taking technology or hardware that cannot be mass-produced to develop the system will eventually lead to the entire system not being mass-produced.
Under the guidance of these two principles, Lingjun Technology first designs the decision planning system, and then puts forward the requirements for the perception system on this basis, and then determines its own perception hardware configuration according to the requirements.
As a result, Lingjun Technology has a series of "abnormal operations".
For example, because 360-degree mechanical lidar is too expensive, does not meet the car regulations and affects the shape of the vehicle, Lingjun Technology refuses to use mechanical lidar on the roof, that is, abandons laser SLAM technology and turns to visual SLAM.
▲Lingjun Technology's autonomous driving test vehicle
In the R&D process, Lingjun Technology is the decision-making planning team to the perception team to provide KPIs, the perception team can concentrate limited forces to achieve the most important perception results. In some large autonomous driving companies, the perception team is often at the first place in the research and development process, they provide the perception results with their own understanding, and then hand it over to the decision planning team to make decisions and plans on the existing perception results.
"There is a big problem with the perception-first approach. The perceived results needed for decision-making are often not given, but a large amount of manpower and computing power is used to identify some unwanted goals and results, which is not efficient. Yang Wenli commented to Che Dongxi.
Two figures are enough to verify the results of Lingjun Technology's approach.
First, its autopilot algorithm can be deployed in an embedded controller with only 30 watts of power consumption, indicating that its algorithm is very streamlined.
Second, the entire company of Lingjun Technology has less than 100 people, such a small team can make three kinds of autonomous vehicles into normal operation, and also run well, which shows that its research and development efficiency is very high.
Fifth, this year landed nearly 100 vehicles in 2023 involved in regional logistics
At the end of the exchange, Yang Wenli and Pang Dongjun also introduced the next mass production landing plan of Lingjun Technology to Che Dongxi.
In 2022, its Robobus will be successively put into operation in more cities such as Suzhou, Hangzhou, Nanjing, Wuhan and so on, and it is expected to invest about 60 units.
The first batch of autonomous logistics vehicles built in cooperation with Manbang and Shaanxi Logistics will also be put into test operation, and the total number of buses, logistics vehicles and Robotaxi will reach a total of 100 units, and will bring tens of millions of operating income to the company.
▲Lingjun Technology Beijing Office
After communicating with Yang Wenli for several hours, Che Dongxi deeply felt that the characteristics displayed by Yang Wenli were exactly the characteristics of Lingjun Technology.
As a technical expert, Yang Wenli's first impression is that he is slightly introverted - he does not speak in a loud voice and speaks slowly. But once he talked about autonomous driving technology and industrial development, he was like a different person, with endless technical terms and insights, and from time to time he shrugged his shoulders and talked about the industry, which was enough to see that he loved the automatic driving industry deeply.
The same is true of Lingjun Technology. It seems that "its appearance is not proud" and "its fame is not big", but it has great potential - it has put forward innovative and pragmatic new ideas in technology research and development, and has achieved mass production of L4 autonomous driving products with very little capital and team, and can firmly adhere to its own development path and rhythm.
To sum up, although Lingjun Technology was not in the center of the stage before, it is possible to achieve a counterattack in the marathon of automatic driving.
After saying goodbye to the car, Yang Wenli walked to the office across the street in the dark.