Although the wave of intelligence is reshaping the entire automobile industry chain, although electric vehicles are an excellent entry point for outsiders in the automotive industry, the new forces represented by Xiaopeng Automobile still regard smart driving as the core of traditional car companies. Differences. What do new car companies do compared to traditional car companies' layout and exploration in intelligent driving, including autonomous driving? What is the difference between them and traditional car companies?
Now do a good job of L2.5 level
'AI+ Internet + electric car' is a huge opportunity in this era. The main founder of Xiaopeng Automobile discovered this opportunity as early as 2014. Just then, Xiaopeng Automobile was officially born. Now, find this opportunity. The number of companies and founding teams has increased, including the traditional car giant is also welcoming this change.
Differentiation development is one of the important issues that a new car company must consider. Xiaopeng Automobile co-founder and president Xia Wei clearly stated at the 2018 Future Auto Show and Future Car Developers Conference: 'We want to be in artificial intelligence and Finding our differentiation in the combination of cars, we create high-value, high-quality, highly intelligent evolution of AICar for young users.
Xiaopeng Auto has always positioned its products as smart cars. AI Car must have the ability to perceive, think, learn from each other and communicate with the environment. Xiaopeng Automobile Chairman He Xiaopeng mentioned more than once that the core of Internet cars lies in AI capabilities. Scalable, safely evolving autopilot and Internet capabilities are one of the key foundations of smart cars. But how to integrate smart hardware and software design with a car and upgrade, but the current industry is facing No small challenge, Xiaopeng Automobile saw the opportunity to make real AI cars in the future.
At present, the industry's national road traffic safety administration (NHTSA) (Level 0-Level 4) or the American Society of Automotive Engineers (SAE) (Level 0-Level 5) grading standards for smart cars to determine the degree of intelligence of the vehicle. The higher the level, the more advanced the technology is. But in the case of Xiaopeng Automobile, what is currently being implemented is only the automatic driving of L2.5 in China.
He Xiaopeng demonstrated the automatic parking function of Xiaopeng G3 on many occasions. He believes that if an intelligent function can solve the problem of no more than 5%, and solve the user usage scenario of no more than 10%, it means that the user uses frequency. The degree of use and inertia is very low. Therefore, the goal of Xiaopeng Automobile is to make a function not only useful, but also to make it easy for users to use, to use, 'I want to solve a problem, see Xiaopeng Auto's automatic parking function in the future. Is it possible to solve 74% of user scenarios according to design goals. '
In order to achieve this goal, Xiaopeng has made a lot of attempts and efforts. Taking the products that will be delivered in batches as an example, Xiaopeng Automobile G3 is equipped with dual current camera, binocular side camera, four looking camera, three millimeters. Waver, 12 ultrasonic sensors and two in-cabin smart cameras, a total of 25 smart sensing devices. The roof is also equipped with a camera that can record surrounding images at 360°. G3 also provides OTA cloud upgrade service, support for remote download and upgrade Peng Motor officially verified various applications, through the exclusive APP remote control of doors and windows, air conditioning and vehicle monitoring and early warning.
According to He Xiaopeng, there are 20 sensors on Xiaopeng's products for automatic parking. If you need automatic driving, you need more sensors. Xiaopeng's automatic parking function can not only realize vertical parking, but also side. Parking at the side, parking in the oblique position, whether there is a parking line in the parking lot, whether there is a GPS signal in the underground parking lot, I want to be able to solve nearly 80% of the scene, and can completely stop automatically. In addition to this irregular scene positioning, Xiaopeng The car's automatic parking function also hopes to increase the memory and summoning function, so that the vehicle can deeply understand the parking space, such as the user's favorite parking space, the favorite parking track and so on.
Prepare for the 2020 L3 class
By 2020, Xiaopeng Automobile will realize the mass production of L3 self-driving cars. This rhythm is roughly equivalent to the L3 production schedule of some traditional car companies.
It is understood that in order to implement this goal, on June 23, Xiaopeng Automobile and Desai Xiwei signed a strategic cooperation agreement, which is intended to jointly develop the L3 level automatic driving system. Since then, Xiaopeng Automobile will become the first domestic booking The latest host of the new automatic driving domain controller of the company. The data shows that the controller hardware is equipped with the world's top autopilot chip, with world-leading computing and processing capabilities, providing the development of Xiaopeng auto-driving algorithms and systems. On the basis of this, it integrates the self-developed Chinese-style automatic driving software system. The automatic driving control system scheduled by Xiaopeng Automobile adopts the latest released AI supercomputing chip in 2018, which is said to be advanced automatic. Driving a super brain, it can fuse the sensory data from sensors such as vehicle radar, camera, lidar and ultrasonic, and is designed to meet strict vehicle standards such as ISO26262 road vehicle functional safety regulations. The L3-class automatic driving system will be cooperated. Provide low-speed valet parking, medium-speed traffic jam (Traffic Jam Pilot) and high-speed drive for the domestic Driving intelligent features environmental point of pain.
In the eyes of Desaixiwei, Xiaopeng Automobile is one of the largest OEMs in China's autopilot R&D. What He Xiaopeng values is that the autopilot system provided by Desaixi is equipped with the world's top AI chips. It is expected that the goal of mass production of L3 autonomous driving will be achieved in 2020, and the foundation for mass production of high-intelligent Internet vehicles with six-dimensional sensing capability will be laid.
Not only a partner, Xiaopeng Motor has established production and R&D centers in Silicon Valley, Guangzhou, Beijing, Shanghai, Zhengzhou and other places to implement a global enterprise layout. In terms of talent reserve, Xiaopeng Automobile is in intelligent network and automatic. The speed of 'enter people' in the driving field is also accelerating. It is understood that Xiaopeng Automobile has a team of nearly 2,000 people, and it is expected to be close to 3,000 by the end of the year, and 35% of them are doing Internet and autonomous driving research and development. In the next two years, the R&D team with more than 600 people will cooperate with many world-class companies to develop. By 2019, Xiaopeng will build a team of more than 5,000 people, nearly half of which is doing Internet, auto-driving and service operations. personnel.
Thinking about building AI Car
'With the empowerment of AI, our Xiaopeng car needs to have a car that is suitable for the automatic driving ability of Chinese cities. Xiaopeng Automobile is now developing synchronously from L2 to L3 to L4. We call the process of L2-L3. For L2.5, our system will gradually evolve to L3, L4. 'Xia Wei revealed some important thoughts of Xiaopeng Automobile on building AI Car.
Xiaopeng Auto has always positioned its products as smart cars. AI Car must have the ability to perceive, think, learn from each other and communicate with the environment. Xiaopeng Automobile Chairman He Xiaopeng mentioned more than once that the core of Internet cars lies in AI capabilities. Upgradeable, safely evolved autonomous driving and Internet capabilities are one of the important foundations of smart cars.
First of all, Xiaopeng will use high-performance chips + independent innovation software development. He said: 'To develop autonomous driving, high-performance chips must be used to achieve large computing demands. In this regard, Xiaopeng has developed its own super The computing platform will be announced at the right time. '
Secondly, Xiaopeng Automobile attaches great importance to the massive Chinese scene big data. In view of the different road environment in China and the usage habits of Chinese users, including China's traffic signs (lane lines, traffic lights, etc.) and foreign countries, there is no small difference, In this process, how to use the massive Chinese scene big data to learn, optimize and upgrade the evolution of Xiaopeng Auto's automatic driving system is a very important direction. Xia Yu further explained: 'This is why Xiaopeng Automobile is the first to deliver internal Employee vehicles, in order to obtain a large amount of data for analysis and development.'
The third is AI Car's 'six-dimensional' perception. For smart cars, perception is very important. It includes self-perception, precise positioning perception and predictive perception.
Xiaopeng's products have been equipped with voice control, real-time traffic navigation, mobile remote control, OTA remote upgrade and other functions. AI is not only used for autonomous driving, but also will give Xiaopeng a closed-loop integration of intelligent network connectivity. In addition, small Peng Automotive is still in the development process, and introduced AI in manufacturing.
There is also the system's security design. Whether it is unsafe to develop a networked car is a problem that many people are worried about. How to dispel such concerns, Xia Wei said, 'Many car companies rarely have top cyber security experts, and Xiaopeng From Huawei, Xiaomi has recruited many top cyber security experts, including special security chips in our next-generation models. This is different from many car manufacturers. In addition, we employ a lot of hackers from various Dimensions, attacking hundreds of trial-running vehicles from all angles, so far, our cars have guaranteed our network security very beautifully.'
How to build autonomous driving with Chinese characteristics
In the overall design concept, according to Gu Junli, vice president of R&D, Xiaopeng Automobile, the intelligent function of Xiaopeng’s first production car G3 adopts integrated + independent research and development program, which integrates some of the existing technologies. On top of the study of AI, the function is extended. The advantage of this strategy is that the technical risk of the initial stage of the car is limited.
But in the second generation of cars, Xiaopeng Automobile will be more daring to liberate ideas, adopting the most advanced chips in the world, and independently researching and developing chips and software to realize the landing of the auto-driving Chinese scene.
Auto-driving with Chinese characteristics is the goal repeatedly mentioned by Gu Junli since she took office. She believes that the core of autopilot is algorithm, the algorithm comes from data, and the data comes from the scene. To solve the problem of auto-driving in Asia and even China, we must Starting from the scene, collecting a large amount of driving big data from the scene, and then evolving, improving, innovating, and solving more and more problems in the driving scene data, and continuously performing real-time calculation in high-performance chips.
Because Asian driving scenes have their own unique density and complexity, Gu Junli said, 'This determines that we need to use a custom sensor to structure more sensors, higher density sensors to solve this problem. Combine driving scene data, combined AI, combined with high-performance chip and sensor architecture, solves the problem of auto-driving in China and even Asia. It requires a vertical software and hardware system, which is also a vertical system that Xiaopeng will focus on.
At the sensor level, multi-dimensional camera, multi-dimensional radar, ultrasonic and other sensors for heterogeneous multi-sensor fusion, can achieve multi-scene, full range, no dead angle coverage, Gu Junli pointed out that this is necessary to drive in Asia If you want to have it, such as encountering a gaze, touching a porcelain, etc., the perception of the vehicle required by many domestic scenes is stronger.
It is understood that there are more than 10 sensors on the Xiaopeng G3, and more than 30 sensors will be used in the next generation of vehicles. With these sensors, it is also necessary to make complex scenes and advanced decisions through AI. Gu Junli proposed: 'We Driving ability does not have to rely on the cloud or the Internet, but rely on our own AI, high-performance computing chip, real-time interaction with the local car, can achieve their own driving function, this is the individual intelligence of our products. Our cars are connected to other vehicles and connected to the cloud to get more information. This information can be used for self-learning to get community intelligence.