"How to safely land on autonomous driving" has become the focus and consensus of the intelligent networked automobile industry.

　　 On September 26, Haomo Zhixing CEO Gu Weihao delivered a keynote speech on "Autonomous Driving Safety First" at the 2021 World Intelligent Connected Vehicle Conference. At present, autonomous driving is not only an important engine for the transformation and upgrading of the automobile industry, safety also brings unprecedented attention to the development of the industry, which is the "first lesson" that the entire industry must face. Gu Weihao believes that the safety of autonomous driving is a systematic project, and product capabilities, correct use of users, and policy guidelines are indispensable. "Autonomous driving companies must first develop their internal skills for ‘safety’. Haomo Zhixing was born out of the OEM, and safety is a consciousness engraved in the bones. We also believe that ‘safety’ should become the DNA of every autonomous driving enterprise.”

　　 (Haomo Zhixing CEO Gu Weihao delivered a speech at the 2021 World Intelligent Connected Vehicle Conference)

　　 Gu Weihao revealed that at present, the industry generally uses 1 million pixel cameras, and Hao Mo Zhi behavior guarantees "farther and clearer" safe driving. In 2022, it will take the lead in mass production using 8 million pixel cameras. At the same time, the technology and measures such as China's first CSS safety awareness model and cloud security personnel developed by the full stack are all adding to the safe autonomous driving in multiple scenarios.

　　 At the World Internet Wuzhen Summit held on the same day, Tesla CEO Elon Musk also said: Data security is the key to the success of smart connected cars. Tesla is pleased to see the introduction of relevant laws and regulations to strengthen data management. Gu Weihao and Musk jointly voiced the issue of "safety" of intelligent networked vehicles at two top summits in North and South.

This time, Hao Mo Zhi Xing CEO Gu Weihao was invited to attend the "Shunyi Special Session on Autonomous Driving Applications in Urban Complex Scenes". Xia Huaxia, vice president of Xidian and Meituan, jointly discussed the "safe landing of autonomous driving."

　　 (Haomo Zhixing's low-speed logistics vehicle "Little Magic Camel Logistics Edition" will be unveiled at the 2021 World Intelligent Connected Vehicle Conference)

The World Intelligent Connected Vehicles Conference is co-sponsored by the Beijing Municipal Government, the Ministry of Industry and Information Technology, the Ministry of Public Security, the Ministry of Transport, and the China Association for Science and Technology. International conference.

　　 The following is a record of Gu Weihao’s speech:

　　Hello everyone, I’m Gu Weihao, CEO of Haomo Zhixing. Today, I am honored to share with you some of Haomo’s core thinking and achievements in autonomous driving safety. First of all, I would like to introduce the company. Haomo Zhixing was registered and established in Beijing in 2019. It is an artificial intelligence technology company dedicated to autonomous driving. The company is driven by data intelligence, taking advantage of cost and scale to gradually implement autonomous driving products. At present, it has developed three major businesses: passenger cars, unmanned logistics vehicles, and intelligent hardware, and has achieved initial results.

　　 After more than two years of hard work, we have now mass-produced ten products, including passenger cars on WEY Mocha, tank 300 and other models. It is estimated that the number of passenger cars will exceed 1 million in the next three years. I also thank Miss Xia (Vice President of Meituan and President of Smart Transportation Platform Xia Huaxia) for her trust and cooperation with Meituan for more than a year, and also thank the leaders of Shunyi District for supporting our unmanned logistics vehicles under everyone’s supervision and encouragement. Actively innovate.

　　 Although there seem to be many products, we have a very basic underlying logic for all products, which is security.

　　 We believe that the safety of autonomous driving is a system engineering. Product capabilities, correct use of users, and policy guidelines are indispensable. Security requires every link to be designed and implemented as the most important underlying logic. The product design stage must be complete, and the R&D implementation must be thoroughly implemented. On the vehicle side, there must be redundant and sensitive sensing equipment, a fast brain, and a consistent reasoning system.

The iteration of artificial intelligence requires more data feedback and data training. In this process, it is necessary to do a good job of data security, use the computing power of the server to do more sufficient security verification, at this stage, or in a long period of time, it is also The security of the cloud is required. As the last gate of security, a stable, low-latency 5g is required, and information security is also required to ensure the accuracy and security of control. Of course, the most important thing is that all product implementation practices need to be carried out under the guidance of policies.

　　 Below, I will briefly introduce the practice of Haomo in each link.

　　 First of all, in terms of security design, Haomo Zhixing built a model of collaborative scene security. We believe that assisted driving, autonomous driving, and manual driving will all appear together in the social traffic flow for a long time. Our goal is autonomous driving or assisted driving, which can reduce the traffic accident rate of the entire society. Therefore, we have summarized some axioms that the program must comply with during driving, as the criterion for the realization of automatic driving, and then use symbolic learning or deep learning to achieve. In terms of symbol learning, we have designed a scene library to require every vehicle's actions in the same scene to be interpretable; deep learning and reinforcement learning are used to increase the generalization of driving behavior. In this way, the vehicle becomes smarter and smarter.

　　Secondly, we also designed car-level functional safety. Three levels of safety monitoring have been set up for autonomous driving or assisted driving systems. The first layer monitors whether the software function on the AI ​​chip is normal, the second layer monitors whether the AI ​​chip is working properly on a highly reliable chip, and the third layer monitors whether the main control is working properly on another independent hot standby controller. Use these three levels of monitoring to ensure the safety of the vehicle.

　　Another safety design is based entirely on prevention. We call it anticipatory functional safety. The core idea is to clarify the operating scenario of the system in the design stage, and improve the security of the system through functional degradation and user participation.

　　 Finally, the safety design practice is our dual reasoning system. There is a book called "Thinking Fast and Slow", which shows that there are two decision-making systems in the human body. One is subconscious, reacts extremely fast, and can respond quickly to external stimuli; the other is a causal system, which is slower. The two systems are used in cooperation to express who we are today.

　　In the field of autonomous driving, this idea is also very important. We know that today's deep learning is not very interpretable, but it is fast and very generalized. Therefore, we designed a dual reasoning system, hoping to find more solutions with the deep learning system, and use the symbolic reasoning system for verification, so as to ensure the consistency of the CSS just mentioned.

　　 No matter how good the design is, it also needs to land. In Haomo, we designed and implemented the 3S R&D process, which we call safe, stable and efficient. The development process has designed 23 check points and more than 30 check tools to ensure that every function, every pair and safety design can be implemented.

On the car side, we are also leading the use of more new sensors. Today, the mainstream is still a 1-megapixel camera. We will take the lead in mass production next year to use an 8-megapixel camera, so that we can see and recognize objects farther. , So that the system has more time to prepare. Of course, we will also add lidar and use it in conjunction with the camera to improve the security of perception.

　　 Simulation verification is today an important part of all autonomous driving safety guarantees. In the traditional simulation verification, the tester takes time and effort to configure the simulation scene, and does a lot of random and repetitive work, which is not efficient. We have also innovatively transformed the semantic scenes defined in the CSS cognitive model and automatically carried out simulation scene conversions. Through parallel simulation in the cloud, we have greatly improved the verification efficiency, helped us find problems in advance, and ensure the safety of landing.

　　 In the recent practice of unmanned delivery vehicles, we have also found that cloud security officers are indispensable, and they have become an essential part of today's autonomous driving systems. They can prevent the occurrence of safety risks in advance and serve as the last line of defense for operational safety.

Just now Director Liu (Shunyi District Economic and Information Bureau Deputy Director Liu Jianxin) also issued the "Guidelines for the Implementation of Unmanned Distribution Management in Shunyi District". This guide was released very timely, including open roads, opening hours, and open standards. A detailed rule. I also believe that with such a government team to help the rapid construction and healthy development of the autonomous driving industry can better serve the transportation power of the country, I believe that we will usher in a new situation and new format of unmanned delivery sooner. Welcome and embrace the new form of unmanned delivery together.