Horizon software and hardware synergy leads end-to-end intelligent driving

Ray Technology

2024-05-17 11:41Posted in Guangdong science and technology creators

The Zhiji L6 will be launched, the WEY Mocha Hi4 model will be officially launched, the all-new Changan CS75 will be launched in the second half of the year, the Nezha L pure electric version is expected to be launched in June, and the AVATR 07 will be launched within the year...... The automotive industry will usher in a large wave of new cars entering the market, and new energy vehicles have entered a period of rapid development and continuous involution.

The first half of the transformation of the automobile industry is electrification, the second half is intelligence, and the focus of industry development has gradually shifted from simple new energy technology to higher-level intelligent driving technology. From the just-concluded Beijing Auto Show, it can be seen that consumer demand has changed, intelligent configuration has become an important measurement factor for car purchases, and the penetration rate of intelligent driving continues to rise, which means that the main goal of the market in 2024 will be to achieve the wide application and large-scale implementation of high-end intelligent driving technology.

Urban NOAs are not yet fully "usable" in the eyes of users

Although the high-end intelligent driving technology represented by urban NOA can theoretically bring many benefits to drivers, from the feedback of the market, many automobile manufacturers are still in the exploration stage in the application of urban NOA technology, and it is still difficult to fully meet the intelligent driving expectations of consumers. Whether it is the limitations of the applicable scenarios or the poor experience during driving, these technologies have not yet achieved "usable" and "easy to use" in the eyes of users.

The industry generally believes that the intelligent driving system will go through three stages of development from usable, easy to use to easy to use, respectively, to meet the physical indicators, to achieve anthropomorphic experience, to achieve intelligent driving equality to distinguish the key cognition of each stage.

Horizon software and hardware synergy leads end-to-end intelligent driving

Dr. Kai Yu, Founder and CEO of Horizon

Dr. Kai Yu, founder and CEO of Horizon, once proposed: "A good intelligent driving system should provide an anthropomorphic driving experience, and its system behavior needs to meet the user's psychological expectations as a driver or passenger and all other surrounding traffic participants." "Only when it meets the user's expectations can people gain a sense of psychological security, achieve a more elegant and calm driving experience that can be activated at any time, and promote the user's cognition of the intelligent driving system to continue to be easy to use. Then, through technology iteration and upgrading, cost optimization, and efficient application and mass production efficiency, the high-end intelligent driving experience can be enjoyed by everyone, and finally realize that every consumer truly "loves to use".

Judging from the current popularization status, many automotive intelligent driving systems have not yet entered the use stage of "user satisfaction". Users still maintain a high degree of vigilance for the safety of the system, remain alert to the possible accidents of the intelligent driving system, and hold a wait-and-see attitude, and it will take time to achieve the ultimate goal of "deep love" and "frequent use" of the system.

The end-to-end system is the key to driving the urban NOA towards "easy to use and easy to use".

Driving itself is a group behavior, and each driver has his own expectations and habits, which means that if intelligent driving really wants to be safe and efficient, he must drive like other old drivers on the road, interact with each other, and draw inferences. Therefore, a user-friendly intelligent driving system must have an anthropomorphic driving experience.

So how do you anthropomorphically respond to accidents and unknown situations while driving?

Perhaps only technological reconstruction can more effectively realize the mass production of high-end intelligent driving systems in all scenarios. After the 2023 Tesla FSD V12 version demonstrated the effect of the end-to-end model on the car, the industry generally recognized this development trend: a more anthropomorphic end-to-end system will be closer to the way of thinking of people, directly from input data to output results.

This means that a larger part of the system is built with data-driven modules, and the proportion of modules maintained by humans is reduced, making the overall system easier to maintain. In addition, the end-to-end architecture realizes the functions of multiple models through one model, and R&D personnel only need to train, adjust and optimize the overall model to achieve performance improvement, so they can better concentrate resources and achieve function focus.

As early as 2016, it took the lead in proposing the end-to-end evolution concept of autonomous driving, and in 2022, it proposed Sparse4D, an end-to-end algorithm for autonomous driving perception. Last year, UniAD, the industry's first publicly available end-to-end autonomous driving model by Horizon Robotics, won the best paper in CVPR 2023.

At the Horizon Intelligent Driving Technology Product Launch Conference on April 24, 2024, SuperDrive, a mass-produced full-scene intelligent driving solution based on the end-to-end technology developed based on the flagship version of Journey 6, was released, creating the best high-end intelligent driving system combining software and hardware, and is committed to accelerating the large-scale mass production of the easy-to-use urban NOA solution.

This SuperDrive is a breakthrough in anthropomorphic experience, with dynamic, static, OCC (Occupancy Occupancy Network) three-in-one end-to-end perception architecture, and data-driven interactive game algorithms, it can take into account the scene pass rate, traffic efficiency and behavior anthropomorphism in any road environment, even in complex urban scenarios such as congestion confluence, intersection interaction - dynamic Driveline, courtesy to cyclists, congestion lane change, and urban roundabout. It can also bring an elegant, calm and confident intelligent driving experience.

Under this architecture, the occlusion recall rate can be increased by 70%, the number of dynamic code lines will be reduced by 90%, and the network load can be reduced by 50%, which can effectively solve the problems of high latency, multiple rules, and heavy load in the current industry-aware architecture. The data-driven interactive game can bring a more anthropomorphic optimal solution, making SuperDrive flexibly handle complex traffic flows like an old driver, increasing the success rate of lane change by 50% and the intersection pass rate by 67% in congestion scenarios. Achieve lower maintenance costs while improving computing efficiency.

With an end-to-end system to solve the problem of mass production and landing of NOA in urban areas, Horizon has taken the lead in making a better solution for the development of the domestic high-end intelligent driving field.

Hardware and software collaboration to achieve the best end-to-end mass production efficiency

In addition to mastering cutting-edge software algorithm technology, it is also necessary to build a strong hardware infrastructure and engineering large-scale production capacity, give full play to the synergy between hardware and software, and create extreme efficiency with extreme openness, so as to accelerate the arrival of the era of intelligent driving equality.

Relying on the hardware technology and software technology accumulated over the years as the core support, Horizon provides the optimal path to realize the mass production efficiency of high-end intelligent driving with the full-stack technology concept of "combination of software and hardware".

At the hardware level, Horizon's new generation of on-board computing solutions, the Journey ®6 series, has a unified hardware architecture, a unified tool chain and a unified software stack, and independently designs and develops the intelligent computing architecture "BPU Nash", which natively supports the Transformer model with large parameters, and provides the strongest Transformer computing performance in the industry, providing strong support for the deployment of advanced end-to-end systems, and bringing higher mass production efficiency in cooperation with car companies and Tier 1 supplierswhile reducing costs.

Among them, the journey 6P for the high-end urban intelligent driving market is the culmination of advanced processor technology. A single Journey 6P can support full-stack computing tasks such as perception, planning and decision-making, and control, and at the same time achieve high cost performance under the premise of high performance, which greatly reduces the cost of hardware systems.

At present, the Horizon Journey family has shipped more than 5 million pieces, reached pre-assembly mass production with more than 30 car companies around the world, and cooperated with more than 110 mass production models, with rich mass production experience and solid engineering large-scale production capacity.

At the software level, SuperDrive™, a full-scene intelligent driving solution, realizes accurate perception and intelligent decision-making of complex road conditions in urban areas. Through close cooperation with hardware, the software algorithm can give full play to the hardware performance, realize efficient data processing and computing, and provide stable and reliable software support for the NOA system.

The key to the synergy between software and hardware lies in the seamless connection and efficient cooperation between the two. SuperDrive and Journey 6P create the best high-end intelligent driving system that combines software and hardware, so that the easy-to-use urban NOA solution can accelerate the large-scale mass production and realize the reach of everyone. By continuously optimizing the interface and communication protocol between software and hardware, Horizon Robotics realizes efficient data transmission and collaboration between hardware and software. This not only improves the overall performance of the NOA system, but also reduces system power consumption and cost, laying a solid foundation for optimal mass production efficiency.

On this basis, Horizon Robotics has also continuously polished and improved the software and hardware performance of the NOA system through large-scale testing and verification work. Through close cooperation with major car companies and partners, Horizon Robotics has successfully applied the NOA system to a number of mass-produced models, accelerating the mass production of NOA in urban areas.

At the Horizon Intelligent Driving Technology product launch conference, JOYNEXT and ZeroBeam Technology have announced that they have reached a cooperation with Horizon Robotics to jointly create a mass-produced SuperDrive domain controller solution. At the same time, Horizon also announced that SuperDrive will reach cooperation with a number of top Tier1 and automotive brands in the second quarter of 2024, and will launch a standard version of the production plan in the fourth quarter, and will achieve the delivery of the first production cooperation model in the third quarter of 2025.

I believe that this set of high-end intelligent driving "play" combining software and hardware will provide valuable experience and reference for the development of the entire industry, and I also look forward to the fact that in the near future, more people will be able to enjoy the convenience of travel brought by the popularization of high-end technology, and truly return technology to life, so that everyone loves to use it and is easy to use everywhere.

View original image 92K