"The main competitors of L4 autonomous driving companies in the future will be car companies represented by new car-making forces."
In the past, autonomous driving companies were often regarded as suppliers of car companies, but with the transformation and development of the industry, some L4 level autonomous driving companies have also begun to enter the field of car manufacturing and stand at the center of the industrial chain.
At the same time, a new consensus has gradually formed in the industry that L4 level autonomous driving technology will take the lead in achieving commercial mass production in truck logistics scenarios.
In this context, Former Vice President of Xiaoma Zhixing Zhao Ruixuan joined hands with Waymo Technology Bull Wang Qingzhou to establish Xingjiao Technology in August 2021, focusing on the commercial landing of L4 level autonomous electric truck logistics.
Xingjiao Technology defines itself as a new force in trucks and adheres to the development route of building electric vehicles + self-developed automatic driving. Its car manufacturing method is similar to the D1 jointly created by Didi and BYD, and the ultimate goal is not to sell the car, but to carry out logistics operations.
At the beginning of its establishment, Xingjiao Technology completed millions of dollars in financing. Eight months after its establishment, Xingjiao Technology and the car company released an L4 level autonomous pure electric van Apebot I on the product side.
This model uses Xingjiao Technology's electric-by-wire chassis and L4 level autonomous driving front-loading mass production plan, and is expected to achieve mass production early next year.
01
Truck + logistics, a bigger market than passenger car operations
According to the scenario division, there are three trillion markets for autonomous driving:
Unmanned rental (Robotaxi)
Robotruck
Smart Car Tier1
Among them, the main application of unmanned trucks in addition to ports, mining area scenarios, the largest market is road transport.
Xingjiao Technology mainly focuses on the road logistics scenario, which is a 10 trillion scale market.
Choosing to take the field of autonomous truck logistics as the entry point, Zhao Ruixuan told us three reasons:
First of all, from an operational point of view, logistics is a larger market than passenger ride-hailing.
Anxin Securities has calculated that the total mileage of domestic passengers is expected to reach 10 trillion kilometers in 2030, and the penetration rate of Robotaxi is expected to reach 6% to 22%, with a unit price of 2 yuan / km.
Taken together, Robotaxi's mobility market size is expected to reach 1.2 trillion to 4.4 trillion yuan in 2030.
In contrast, according to the Billion Euro think tank, the size of China's road freight market will reach 5.85 trillion yuan in 2021 alone.
According to data from the National Bureau of Statistics, the total cost of social logistics in China in 2020 is 14.9 trillion yuan, of which transportation costs account for 52%, as a major country of road freight, China's road freight volume accounts for more than 75% of the total social freight volume for a long time.
At the same time, relevant data show that in 2020, the country has 11.7154 million road operating cars, 11.1028 million trucks, and 157.8417 million tons. Among them, there are 4,141,400 ordinary trucks, 506,700 special trucks, 3,108,400 tractors, and 3,346,300 trailers.
Generally speaking, domestic trucks are basically operating vehicles, which are very sensitive to cost and need to reduce costs and increase efficiency through automatic driving.
Secondly, the electrification of trucks is slightly later than that of passenger cars, which is a good entry point.
One of the core ideas that Xingjiao Technology adheres to is that autonomous driving must be on the platform of electric vehicles. Electric vehicles have obvious advantages over fuel vehicles in terms of the underlying electrical energy layout, network communication, electronic and electrical architecture, or linear control.
At present, the electric trucks of some companies in Europe and the United States have achieved small-batch mass production or listing stages, but the development of domestic truck electrification is relatively slow, which is a very good window opportunity for start-ups.
Finally, in the L4 level autonomous driving scenario, freight logistics will land faster than passenger cars.
High-speed logistics is a semi-closed scene with no traffic lights, mixed traffic of people and vehicles, and complex intersection conflict points. High-speed scenarios will not face the large number of urban complex long-tail corner cases that Robotaxi encounters today.
Even in slightly more complex scenarios in urban areas, the route path of freight is relatively fixed, which makes the algorithm design not particularly complicated by too many long-tail problems, and better supports the data-driven algorithm architecture and the step-by-step landing method.
The scale of the freight logistics market is large enough to commercialize faster than Robotaxi, which also makes many players focus on the track, including Waymo, Aurora, Baidu, Xiaoma Zhixing, Didi, etc. have begun to cut into the trunk logistics market.
Compared with these players, although Xingjiao Technology is a newly established start-up company, it has cut into car manufacturing from the beginning of its establishment, although it is a big challenge, but it is a market that has a larger space than simply doing autonomous driving imagination.
02
Car building + automatic driving, to achieve software and hardware integration optimization
The entire business development model of Xinglan Technology can be simply understood as car building + self-developed automatic driving system, forming its own logistics operation fleet.
Why build a car?
In Zhao Ruixuan's words, it is to define the hardware yourself, which can help the software to be better optimized.
Alan Kay once said that people who really write software seriously should make their own hardware. This sentence was carried forward by Jobs at the first-generation iPhone conference in 2007.
Google only wanted to do the Android system to adapt to all mobile phones, but every time the function was updated, it needed to adapt to all models, which undoubtedly added a lot of difficulties to the team.
In contrast, Apple only needs to adapt its own hardware devices, but can optimize the software system better.
The structure of the parts of the car itself is more complex than that of the mobile phone, and the automatic driving system needs to be deeply adapted to the body to achieve the control of the whole vehicle body.
Therefore, it is difficult for the automatic driving system to be completely universal, even if it is a unicorn company, it can only be deeply bound on one or two cars in the end to better optimize the algorithm. Even so, technical compromises are often required to match the product planning of car companies and Tier 1.
In addition, the biggest difference in the structure of electric vehicles and fuel vehicles is the power system and energy supply system, electric vehicles use batteries, motors, controllers and related equipment to replace the original internal combustion engine and fuel tank.
Generally speaking, the number of parts for ordinary fuel vehicles is about 30,000, and the electrification of automobiles will reduce the number of parts by half. This makes the product standardization of electric vehicles higher and the industry entry threshold lower.
This also makes the difficulty coefficient of Xingang Technology drop by an order of magnitude when defining the demand for automotive products.
In terms of car manufacturing methods, Xingjiao Technology chose to cooperate with foundries, but the former has full control over the core three-electric system, the definition of the wire control and computing platform, the automatic driving system and the software.
Xingjiao Technology has launched the first production car, the Apebot I:
This is a 4x2 van, the total weight of the car is 18 tons, the length of the container is 9.6 meters, and the maximum volume of the container is 67 cubic meters, which is currently the most commonly used model across the city in the logistics province.
The car adopts a leading integrated electric drive rear axle solution, and applies a liquid-cooled battery system customized for high-speed working conditions, with an actual full load endurance of 380km, which can meet most of the cross-city transportation needs.
Xingjiao Technology said that apebot I is currently the world's first L4 self-driving front-loading pure electric vehicle model.
Apebot I is currently preparing for various validation tests prior to mass production, with mass production expected in February 2023. This is also the world's first chassis in the true sense of the world to open up electric and wire-controlled.
The self-developed software control algorithm and advanced E/E architecture make the motor and electro-hydraulic steering and EBS system more accurate and smooth, so as to better realize the control of the execution unit by automatic driving.
In order to better balance the research and development of car manufacturing and automatic driving systems, there are two complete and independent teams within Xingjiao Technology:
CEO Ruixuan Zhao was previously vice president of self-driving company Xiaoma Zhixing, former general manager of Facebook's game business in Greater China, and executives in companies such as Sina Weibo and Lenovo Group.
During the period of Xiaoma Zhixing, he completed the commercial cooperation and financing of a number of car manufacturers and Tier 1, including Toyota, and led the truck to achieve the first high-speed logistics scene in China.
Autonomous driving team: Mainly CTO Wang Qingzhou, who is the former technical head of Waymo's architecture department and has served as a long-term technical leader at Google's Machine Learning Research Institute. Graduated from the Department of Automation of Tsinghua University, he entered Google's US headquarters in 2011 and joined the Waymo Architecture Department in 2017, where he was responsible for the research and development of multiple core technologies for unmanned vehicles.
Vehicle team: Mainly Ma Junye, vice president of vehicle engineering, has nearly 15 years of experience in the vehicle research and development industry, and has served as the chassis director of Xiaopeng Automobile and the director of the chassis department of BAIC.
Xinglan Technology has a technology research and development center in Silicon Valley, usa and Shanghai, China, and the team members are from Google, Pony.ai Xiaoma Zhixing, Qualcomm, Xiaopeng Automobile, Bosch, ZF and other industry leaders.
The ultimate goal of Xinglan Technology is to build an efficient unmanned electric transportation network, all autonomous vehicle fleets operate independently, provide extremely costly transportation solutions, and the main customer groups are also major logistics companies or cargo owners.
03
Powered by 3 solid-state lidars
Vision-based multi-sensor fusion system
Zhao Ruixuan believes that the ultimate form of the future truck will be to remove the cab, only an electric chassis equipped with a box, and finally the core difference is reflected in the company's automatic driving capabilities and software and hardware optimization capabilities.
This is also the advantage of Xingang Technology.
Car manufacturing, so that xinglan technology has the ability to independently define hardware, based on many years of industry experience in automatic driving, on the one hand, it quickly defines and develops suitable automatic driving hardware, and combines the vehicle engineering to achieve hardware front-loading solutions.
On the other hand, its autonomous driving system can be adapted to the self-defined hardware system, so as to achieve better optimization iteration.
Specifically, at the hardware level, the L4 self-driving truck system is based on the deep integration of multiple sensors, including 9 cameras, 3 millimeter wave radars, and 3 solid-state lidars, which can cover an area of 250 meters forward and 150 meters backwards.
Of course, the entire system architecture of Xingjiao Technology is not completely dependent on lidar, but more based on vision, and lidar plays a role in safety redundancy.
Here we will focus on the algorithm perception system APANet independently developed by Xiaxia Ape Technology.
APANet is a multi-task neural network architecture that integrates multiple functions such as obstacle detection, lane line detection, front-and-back semantic segmentation, image depth estimation, bird's-eye view object detection, and lane line detection.
The camera in APANet can bring richer detail information to the entire system, through the most advanced algorithm, can directly detect the 3D position of the obstacle of the camera angle, and can implicitly learn the projection relationship from 2D to 3D through the neural network, and integrate the features of multiple cameras in the BEV space to learn the accurate position information of the surrounding environment and obstacles.
Lidar, on the other hand, provides rich location information through physical point clouds, and lidar in APANet can learn accurate obstacle information from multiple perspectives such as front view and top view.
At the same time, after the introduction of graphical semantic information, the features of lidar have better classification ability for pedestrians and small objects, while the visual features can improve the recall of small sample data after the introduction of lidar features.
Most importantly, Xingjiao Technology's APANet perception algorithm does not rely on high-precision maps at all, which can identify landmarks such as lane lines and traffic signs in real time, which can help Xingjiao Technology's positioning system to achieve self-vehicle positioning only by relying on GPS, low-end IMU and visual features without the need for lidar and high-definition map matching.
There are three more benefits to not relying on high-precision maps:
Not limited by HIGH-definition maps, it can be driven in any ODD range;
Reduces the huge cost of HIGH-definition map collection and maintenance;
The on-board system calculation platform does not need to reserve a huge memory space for high-precision maps, and also greatly reduces the requirements for the accuracy of the navigation and positioning system, greatly reduces the cost of the on-board system and sensors, and is more suitable for mass production.
From the car to the automatic driving system, for xingjiao technology, the software iteration and optimization of the automatic driving system is still the key to the future competitive development, and when the hardware is defined, it will have more advantages for the optimization and iteration of its software.
04
Car building, L4 technology landing mass production is the only way?
At present, whether autonomous driving companies want to build cars has always been a hot topic in the industry.
In the early days, Waymo launched a small two-seat prototype, but due to the high cost of passenger car investment, it did not continue its own car-making plan, but turned to ordering models from car companies and modifying its own routes.
At present, including GM's Cruise and Amazon's acquisition of Zoox, have announced their own car manufacturing plans, Didi has also been exposed to independent car manufacturing, the internal car manufacturing business code name "Da Vinci".
Zoox self-driving prototype
Car manufacturing, for L4 level autonomous driving enterprises, is undoubtedly another way of thinking to achieve L4 technology on the car, which can help these companies achieve commercial mass production of L4 technology faster.
Compared with passenger cars, trucks have a much lower investment in vehicle research and development and production lines because of their flexible structure of non-load-bearing bodies and low demand for modeling interiors. In the future, new models without cabs will further reduce the difficulty of building cars.
In general, there are three paths to commercialization of L4 autonomous driving enterprises:
Car manufacturing + self-developed automatic driving system;
Do not build cars, only do automatic driving systems, operate their own fleet;
Do Tier 1 to provide automatic driving systems for car companies;
At present, the L4 level autonomous driving enterprises in the industry that choose to build cars, in addition to Xingjiao Technology, also have Turing smart cards that Tucson will set up in China in the future.
Turing smart truck manufacturing, first of all, will cut from the freight van, and gradually enter the field of heavy trucks, the goal is to create a model that is more in line with the operational requirements of the autonomous driving fleet, the cost is lower, and the experience is better.
Previously, Tucson also cooperated with car companies to develop models, including the earlier cooperation with Navistar, an American heavy-duty truck manufacturer owned by TRATON Group, and the two sides will start mass production of autonomous heavy-duty trucks in 2024.
But I have to say that cooperating with car companies to mass-produce cars, compared with their own cars, the dominance of hardware is not high.
Wincher Technology is based on automatic driving systems, and in 2019, it launched a joint development of autonomous driving heavy truck projects with OEMs, and the two sides also jointly developed a wire-controlled chassis to achieve a fully redundant solution for vehicle-level wire-controlled chassis.
Last year, Wincher Technology showcased two mass-produced models of autonomous driving heavy trucks at the 2021 World Artificial Intelligence Conference, which were jointly developed with Dongfeng Commercial Vehicle and China National Heavy Duty Truck and equipped with the Yingche Xuanyuan automatic driving system.
Suzhou Zhitu Technology is the role of Tier 1, a smart car software and hardware system research and development technology company initiated by FAW Jiefang, which is committed to providing L2-L5 level automatic driving systems for FAW Jiefang and the industry, and empowering the industry with complete solutions for autonomous driving and smart logistics.
In July last year, the J7 super heavy truck for the trunk line scene, which was developed by Zhitu Technology, was officially mass-produced before the production line. At present, Zhitu Technology has realized the intelligent driving landing layout focusing on trunk logistics and covering scenarios such as sanitation, ports, and ports.
At present, L4-level autonomous driving, which is mainly based on operation scenarios, has entered the second half of the competition of commercial mass production, and as the industry's optimistic automatic driving truck trunk logistics scenario that can be commercialized and mass-produced, it has also been competed by many players.
Facing the future, in the three commercialization paths, the route of car-making + self-developed automatic driving system is undoubtedly more imaginative, and relatively speaking, the challenges it will face will be greater.
In the end, who can outperform the end game is worth looking forward to.
"The Heart of the Car · The connoisseur said" trailer
On Thursday, April 21, 20:00-21:00, Box Cars founder and president Zhang Shuguang will be a guest car heart · Experts said, talk about "travel market nuggets: where is the opportunity point of the B-end track of new energy vehicles?"