How to accelerate the commercialization of high-level autonomous driving applications

1. Current status of commercial demonstration and application of autonomous driving technology

In recent years, with breakthrough progress in underlying supporting technologies and software and hardware equipment such as smart chips and smart sensors, autonomous driving has also transitioned from the technology research stage to the product implementation stage.

Developed countries such as the United States, Europe, and Japan have gained advantageous experience in the field of autonomous driving by leveraging their high-end manufacturing capabilities and competitiveness in core technologies. They have quickly launched commercial operation attempts in multiple scenarios such as logistics distribution, low-speed shuttle buses, trunk freight, and manned taxis.

Although my country started a little late in autonomous driving, after 2017, major cities have accelerated the promotion of related work such as testing, verification, and demonstration applications of autonomous driving, and the focus of work has begun to shift from research and development testing to commercial application pilots.

Figure 1 Main technical routes

Commercial application of autonomous driving abroad

The United States is limited by infrastructure conditions, so it focuses on applying autonomous driving to private cars and taxis. For example, Tesla's autonomous driving system Autopilot has been deployed on Tesla vehicles around the world, and is currently promoting "full autonomous driving (FSD)" software upgrades close to the L4 level. It provides consumers with more comprehensive autonomous driving functions through subscription and buyout services.

Waymo, Google's self-driving company, has launched a driverless taxi service in Phoenix, Arizona, USA, and the service area has now reached 466.2 square kilometers. At the same time, it has cooperated with Uber to provide self-driving car online car-hailing services and food delivery services.

Currently, it has achieved more than 10,000 free services to passengers per week, and plans to provide more than 100,000 services to passengers by the summer of 2024.

Figure 2: The Big Three of U.S. Autonomous Driving

European countries focus on promoting the commercialization of autonomous driving around the ecosystem. For example, Finland has developed autonomous minibuses due to its small market and lack of large automobile manufacturers. It relies on its advantages in policies and software technology to create a complete public-private partnership public transportation operation system.

For example, Sweden, based on its advantage in heavy-duty vehicle equipment manufacturing, focuses on the application of autonomous driving in logistics and realizes the operation of truck platoons between cities. Germany, as the first country in the world to legislate on autonomous driving, has begun mass production of truly L3 autonomous driving vehicles and is conducting tests of remote driving services.

As the first country to propose the concept of autonomous driving, Japan attaches great importance to it at the strategic level, but it adopts the strategy of "accumulating small steps to achieve a thousand miles" in terms of tactics. On the one hand, it continuously introduces policies, regulations and technical standards, and on the other hand, it carries out the integrated development of production, learning and research to accumulate technology at the national level.

Its commercial application is mainly applied to short-distance, low-speed shuttle buses due to factors such as the aging of society, traffic congestion in central urban areas, and the public's preference for public transportation. For example, the self-driving golf cart service launched by Miyama City and the guide rail bus service launched by Nagoya City.

Figure 3 Japan's autonomous driving commercial application test

Current status of application demonstration in various cities in China

More than 10 major cities in China, including Beijing, Shanghai, Guangzhou, Shenzhen, Wuhan and Changsha, have launched commercial pilot projects for autonomous driving. Among them, Wuhan, Chongqing, Shenzhen and Shanghai have started to launch fully unmanned commercial operation services. Guangzhou is also simultaneously carrying out unmanned manned tests of autonomous driving to provide technical and operational guarantees for further commercial operation services in the future.

As of April 2023, more than 50 provinces and cities across the country have issued regulations on intelligent connected vehicle testing, with a total of more than 2,000 road test and demonstration application licenses issued, more than 10,000 kilometers of test roads open, a total test mileage of more than 40 million kilometers, and targeted deployment of manned scenario demonstration applications.

Take the progress of Wuhan's autonomous driving commercial pilot as an example. Since the launch of the intelligent connected vehicle commercial pilot in June 2022, the monthly passenger volume of autonomous taxis in the Wuhan demonstration area has increased significantly. By December 2022, the passenger volume of manned taxis exceeded 14,000 people per month, and the passenger volume of fully unmanned taxis exceeded 7,000 people per month.

As of January 2023, the Wuhan Demonstration Zone has opened 522 intelligent connected vehicle test sections in five batches, with a total mileage of 751.56 kilometers and a two-way mileage of 1,503.12 kilometers, covering an area of ​​600 square kilometers in Wuhan City and reaching a permanent population of nearly 2 million, ranking among the top in the country.

In 2023, Dongfeng Yuexiang and Baidu Luobo Kuaipao plan to add 300 autonomous driving vehicles, and the number of vehicles in regular operation will exceed 400. They will gradually establish dedicated autonomous driving operation lines to high-speed rail stations, airports and other core transportation hubs, and provide autonomous driving travel services to more Wuhan citizens.

Figure 4 Schematic diagram of Wuhan autonomous driving car road test area

Changsha also started commercial autonomous driving services relatively early. In June 2019, Changsha issued 45 open road test licenses for intelligent connected vehicles and became the first city in China to implement a trial operation of autonomous taxis.

On April 19, 2020, Changsha's self-driving taxis were fully open to the public, with an operating area covering 130 square kilometers.

In July 2022, the scope of operation of self-driving taxis was further expanded, with a new 317 kilometers of test roads opened. At the same time, new stations were added to connect to public transportation facilities, enabling trial operations in areas with dense passenger flow.

As of August 2023, self-driving taxis have traveled a total of 1.5 million kilometers and served 150,000 passengers.

The application demonstration of autonomous driving in Beijing is being promoted in stages. In April 2021, Beijing established the first intelligent connected vehicle policy pilot zone in China within the 60 square kilometers of Yizhuang, and officially opened the autonomous driving manned testing and commercial operation. The demonstration zone has set up more than 600 drop-off areas, with an average daily total of 3,600 trips, accounting for 1.3% of the total trips in the zone and about 40% of the traditional taxi/online car-hailing trips.

Following Yizhuang, Beijing has opened 52 autonomous driving vehicle test roads in Haidian District, 77 in Shunyi District, and 26 in the sub-center area of ​​Tongzhou District. In July 2023, Beijing launched a commercial pilot project for "no one in the car" for intelligent connected passenger vehicles.

Figure 5 Schematic diagram of Beijing’s autonomous driving open areas

2. Bicycle intelligence and vehicle-road collaboration technology routes are developed in parallel

Based on the application situation at home and abroad, the current development direction of autonomous driving technology is classified on the vehicle side or road side based on the computing power and cost of different solutions, mainly including "single-vehicle intelligence" and "vehicle-road collaboration". Different social entities such as governments, enterprises, and the public take different technical paths for the commercial application of autonomous driving based on different understandings.

The most advanced representative of single-vehicle intelligence is Tesla in the United States. Relying on its core competitiveness in chip research and development and algorithms, it was the first to realize end-to-end large-model application in autonomous driving systems in 2020.

That is, a large AI model is built between road scenes and vehicle control. By capturing the changes in driver operations in different scenes, the machine is trained and trained to continuously imitate and approach the control behavior of human drivers. The use of this shadow mode to achieve rapid accumulation of scenes has helped Tesla quickly achieve coverage of long-tail scenes, widening the gap between it and other autonomous driving companies in scene coverage.

Chinese companies such as Huawei and Xiaopeng have realized the application of large models in vehicle autonomous driving systems in 2023. However, to achieve high-level autonomous driving above L4, the single-vehicle intelligent route must solve the problem of geometric increase and expansion of vehicle-side chip computing power and cloud computing power on the one hand, and on the other hand, it must face endless scene data collection and a large amount of manual identification and labeling of training data.

Vehicle-road collaboration tends to integrate systems, integrating all traffic operation elements into an overall decision-making and planning system. It can not only achieve beyond-line-of-sight, multi-dimensional global perception capabilities, but also optimize traffic safety and traffic efficiency based on broader urban information. At present, the progress of the vehicle-road collaboration solution adopted in my country is still in the collaborative perception stage.