BYD reveals 3 million smart driving cards, Yang Dongsheng: Serious accident rate drops to 1/6 of people driving

On May 21, the 13th Annual Conference on Intelligent Connected Vehicle Technology, co-sponsored by the China Automotive Engineering Society, the National Intelligent Connected Vehicle Innovation Center, the School of Vehicles and Transportation of Tsinghua University, Shanghai International Automobile City (Group) Co., Ltd., and the National Key Laboratory of Intelligent Green Vehicles and Transportation, was grandly opened at Shanghai International Automobile City. At the plenary session, Yang Dongsheng, senior vice president of BYD Group and president of the Automotive New Technology Research Institute, shared the large-scale application data and security achievements of BYD God's Eye Assisted Driving under the theme of "Intelligent Electrical Convergence" to lead the safety upgrade of intelligent vehicles. Yang Dongsheng believes that intelligent technology has spanned from selection to irreversible user demand. BYD has verified the core value of intelligent driving in reducing traffic accidents and relieving novice anxiety through its huge installed base of more than 60 models and nearly 3 million vehicles: the serious accident rate of tens of millions of kilometers has been reduced to one-sixth of that of people driving, and parking scrapes have been reduced to one-fiftieth. Based on this, industrial security is moving to a new stage of "surpassing human security". The integration of hardware, computing power and multi-motor vehicle preparation provides a physical basis for this leap. In the face of the implementation of L2 strong targets in the industry and high-level kinetic energy specifications, BYD is relying on the "Xuanjie Architecture" to open up the electronic architecture and electrification system, and is promoting intelligent driving into the deep water area of "smart electric integration". At the algorithm level, by introducing a physical AI model in the car to predict the future, and introducing a world model in the cloud to cooperate with 190 million kilometers of big data long-tail simulation and reinforcement learning every day, the algorithm can achieve efficient iteration every three days. For extreme parking obstacles such as hollowing out and suspension, BYD achieves accurate identification of dead corners through dual OCC support of vision and lidar, and promotes parking pocket service. Further, in the face of the extreme working conditions of the response lag of fuel vehicles such as puncture tires, ice and snow, the smart-electric fusion development platform achieves extremely fast, accurate and stable control within 200 milliseconds through the digital chassis and motor with millisecond response. In the future, only by combining simulation with systematic measurement and highly integrating "electricity" and "intelligence" in long-tail and extreme scenarios can the industry build a safety base on the boundary just needed by users. The following is a shorthand for the speech: Yang Dongsheng: Experts, colleagues: Good afternoon everyone! It's a great honor to be part of this alliance. BYD We are the practitioners of the first half of electrification and the first half of intelligence. Today, we have the opportunity to share our practices in intelligence and some thoughts on safety. In January 2025, BYD began the comprehensive layout of intelligent assisted driving systems. From 2025 to now, almost all of our car companies have equipped with L2 level assisted driving systems. We have more than 60 models, and the number of our vehicles has also exceeded nearly 3 million. Such a large amount of practice and investment has really made us feel the original intention of intelligent driving. We hope that we can reduce traffic accidents and reduce driver fatigue. We also really heard and felt the feedback of users. It is very easy for everyone to turn on the small blue light to drive all the way, especially many novices who are no longer anxious and busy in the city to see navigation and so on, so we see its value. The consumer activation rate, the activation rate of the pilot function, has reached 50% of the penetration rate, nearly 70% of the parking, and 86%, so it uses a very high viscosity. What value do we see in terms of safety for our users? At the core is the sharp decline in traffic accidents. We use remote viewing to see whether the airbag is open as a reference for serious traffic accidents. Compared to human drivers, it has achieved a level of one-sixth, scratching of parking. We passed the bottom of the parking pocket in July last year, and the consumer usage rate is very high. The scratching of parking has been reduced to nearly one-50th. China's leadership in standards, L2 strong standards will be implemented immediately, L3 and L4 for autonomous driving have also issued drafts for comments. We see that safety is a key attribute of assisted driving and autonomous driving. We see that the industry is standardizing, and BYD is also making active preparations to meet the requirements of standards, which is more to bring safety to users. We have just mentioned superhuman safety. As early as 2024, we proposed that assisted driving and autonomous driving should not create human-like safety, not simply replace humans. It should surpass humans in the future, because its own sensors and computing power, including how many motors are executed on the car, have the basis of surpassing humans. Therefore, BYD has also integrated multiple sensors on the car to create an ultra-safe assistance system. If we want to make a good car, we must look at the whole car as a whole, because most of the systems of BYD cars, especially the core systems, are independently developed by us. We have opened up the electronic architecture system and the electrification system of the whole car. We call it Xuangui Architecture. Only with the support of such a vehicle-level architecture can we have an integrated car that can be considered comprehensively. We have a large amount of data. Through end-to-end models, reinforcement learning, etc., a series of cars are better, the model is better, the motor and the car are electrified, not only the drive motor, the suspension is also the motor, there is already a digital chassis and drive system, so the digital hardware + intelligent upper body, the electric lower body is strong and combined. We also have a good internal development system that can support the key links of our product iteration. Introducing the architecture of intelligent and electrical convergence we launched in 2024, we have integrated the brain, many control algorithms have been moved up, not only intelligent intelligent intelligent driving algorithms, many chassis power driving algorithms have been moved up, through the two ends, cloud and vehicle end capabilities, through network capabilities, and finally through our BYD vertical independent research and development capabilities, our sensing, control, data, mechanical chain is our execution chain, built up. Sensors are still an important dimension that we are constantly improving. Different sensors can solve different perceptual capabilities in the physical world. We all know that lasers, ultrasound, and millimeter waves are good. It has its own exclusive solution for night, occlusion, and so on. Now in the future, we will continue to iterate. We will also be equipped with more advanced sensors to solve this extreme environment such as dazzling, glare, weak light, and so on. In terms of algorithms, we have introduced physical AI models on the vehicle side and the inference side to increase defensive driving ability. At the same time, we have also introduced a world model on the cloud for long-tail simulation. Through reinforcement learning, reinforcement learning is like brushing scrolls, taking college entrance exams, and constantly brushing questions, you can naturally get a good result, because you brush a lot of questions, but the physical AI on the vehicle side is equivalent to staking questions. If you can bet on the problem, because it is predicting the future world, through this reinforcement learning plus future predictions, future algorithms, the model can become stronger and stronger. To solve the continuous improvement of safety capabilities, we need to tap the long tail. Long tail scenarios can generate 190 million kilometers of data a day. Long tail scenarios can be excavated in depth. Intelligent driving algorithms compete for the future iteration capabilities. The competitiveness and iteration capabilities of software are constantly improved. We can do iterations every three days. Simulation mileage estimation is also BYD's closed loop through our big data, sales volume, and data volume. We have also seen that in terms of safety capabilities, AB has triggered almost 9 million times, and the activation rate of assisted parking is very high, which can make users drive and drive safer, reduce accidents, and repair. Like parking, in order to challenge this extreme scenario of parking, we see special, hollowed-out obstacles, as well as dangling obstacles, which are the difficulties of our challenge. These are the ceilings that we constantly mine for algorithmic power and compensate for our parking through data. In order to ensure the bottom of our parking pocket, the user's utilization rate has been further improved. Through the visual OCC and the OCC of the lidar, we have greatly improved the recognition ability, ensuring the bottom of the parking pocket, and the user's loss has been minimized. This is also why we released the Eye of God in July last year. Cars assisting intelligent driving enjoy the permission of this parking pocket. Solving this routine, we believe that more is a kind of automatic driving, assisted driving, everyone is on the routine road to surpass humans, or equal to humans, but in fact, in the future, under the extreme working conditions, the intelligent driving system will be more demanding, so how do we improve safety through the ability of the whole vehicle and the ability of intelligent driving in the extreme working conditions of the early layout, such as high-speed tires, not only BYD, the industry has followed, and how to deal with traffic accidents? These are extremely difficult typical scenarios. Like this rain fog and ice and snow pavement, it is also very difficult to judge the ground capture coefficient. Your light has a strong visual ability, and your driving ability is also stronger. In order to improve the combination of electrification and intelligence, we have created a development platform of smart electricity fusion, which we can achieve for 95 kilometers, completely through the combination of electricity and intelligence. We have also achieved a shorter identification time for high-speed puncture tires, and the maximum stable speed of puncture tires exceeds 200 kilometers. This is not only the ability of intelligence, but also the fast response of the motor, because the motor is a millisecond response. The ordinary engine and fuel truck It is a response of hundreds of milliseconds. Through this transfer amount, the response of almost one second is full, but it can be controlled within 200 milliseconds or even faster in the digital world and the motor world. In order to systematically build our products, we not only have simulation capabilities, we also have actual testing capabilities. BYD We have a special development site for parking alone, and there is also a special development site for driving. Therefore, we can combine simulation and actual testing to create better products, solve our long-tail projects, break through the limits, and bring us a lot of thinking. Next, we hope to discuss with our colleagues the possibility of future scenarios, because we see that the intelligent driver assistance system has now become similar to the era of manual to automatic stopping, it is also irreversible, it has become just what the user needs, it has greatly solved the driver's driving fatigue, but at the same time, it has also raised higher requirements for safety boundaries. We combine electricity and intelligence in long-tail scenarios, so that our products can deliver value when users really need it. Finally, thank you for the invitation. We are also contributing to China's smart driving industry with the support of the alliance. Thank you!