Analysis of autonomous driving landing technology Autonomous driving landing in typical application areas

Preface

As early as 1984, Cameron University in the United States launched the autonomous driving ALV project. Since then, artificial intelligence technology has begun to be tested in the field of autonomous driving, and at the same time, the most basic control strategy for autonomous driving has gradually been formed.

As autonomous driving related technologies become more mature, especially the application of deep learning, and the wave of automobile electrification sweeps in, the autonomous driving wave began in 2015. Led by Google and Baidu in China, autonomous driving startups have sprung up, and traditional automobile OEMs and parts companies have also entered the market.

Then, after several years of R&D and verification, and the capital winter of 2018, people began to return to rationality in their expectations for the implementation of autonomous driving. The 2019 Gartner Emerging Technology Maturity Curve shows that autonomous driving L4 level technology is in the bubble disillusionment period.

Now, with the consensus of the entire upstream and downstream of the autonomous driving industry to increase R&D investment, as well as the improvement of supporting facilities and regulations with government support, the significant reduction in hardware costs and the rapid improvement of software algorithms, the short-term implementation scenarios of autonomous driving are gradually becoming clearer and clearer.

Gartner Emerging Technology Hype Cycle 2019

Analysis of autonomous driving technology

To describe the implementation of autonomous driving, two concepts are indispensable: autonomous driving level and geo-fencing level. Autonomous driving level is used to describe the degree to which autonomous driving can be technically automated; geo-fencing level is used to describe the complexity of the autonomous driving operation scenario.

(1) Levels of autonomous driving

Regarding the level of autonomous driving, the industry currently mostly refers to the SAE definition, which divides the work between the driver and the system when completing the three tasks of driving operations, surrounding monitoring and taking over, as well as the classification of autonomous driving application scenarios.

There are five levels of autonomous driving: assisted driving L1, partial autonomous driving L2, conditional autonomous driving L3, highly autonomous driving L4 and fully autonomous driving L5. As shown in the figure below:

-The first three levels (L1-L3) are the process of continuously expanding the degree of automation of autonomous driving. At the L3 level, the vehicle itself can already achieve automatic control to complete all dynamic driving tasks;

- The last two levels (L4-L5) are the expansion of autonomous driving in application scenarios. L4 can only achieve fully autonomous driving in limited scenarios, while L5 can be expanded to all scenarios. Drivers are no longer required in these two stages;

-L3 level is the transition stage between partial and fully autonomous driving. At this stage, the surrounding driving environment is monitored by the system, but the driver needs to take over the vehicle when necessary.

Automated driving level

(2) Geofence Level

As mentioned above, except for the fully autonomous driving L5, which can adapt to all scenarios, other levels of autonomous driving are operated in limited scenarios. In limited scenarios, the complexity of pedestrian and vehicle flows, as well as the operating speed, can directly reflect the level of autonomous driving capabilities.

Therefore, the concept of geo-fencing is introduced to grade the complexity of limited scenarios from the perspectives of vehicle speed, pedestrian flow and vehicle flow, so as to better reflect the level of autonomous driving. Geographical fencing is also divided into 5 levels, as shown in the following figure:

-Geo1 is the simplest traffic environment (low speed, no people, few vehicles), and the corresponding Geo5 is the most complex traffic environment (random travel, mixed motor vehicles and non-motor vehicles, etc.);

- Geo2 focuses on traffic environments with less traffic (low speed and few people/high speed and no people), while Geo3-4 focuses on traffic environments with heavy traffic

-The main difference between Geo3 and Geo4 is that Geo3 has few people at low speeds and no people at high speeds, while Geo4 has many people at low speeds and few people at high speeds.

Geofence Level

(3) Implementation of autonomous driving

Whether autonomous driving can be successfully implemented depends on two points: first, the autonomous driving level of the implementation plan; second, the geographic fence level of the application scenario.

Obviously, the lower the level of autonomous driving that needs to be achieved and the lower the level of geographic fence of the application scenario, the easier it will be to implement autonomous driving.

The development of autonomous driving capability is essentially a process of continuously expanding to higher levels of autonomous driving and more complex geo-fencing levels. From the perspective of autonomous driving level and geo-fencing level, the current consensus on the current status and trend of autonomous driving is shown in the following figure:

-L1-2 low-level autonomous driving applications are expected to cover all scenarios in the short term;

-Geo1-2 low-level geo-fence scenarios are expected to reach L2 level in the short term to achieve autonomous driving ;

-Geo3 mid-level geo-fencing scenarios, with autonomous driving function expansion in the short term, is expected to be as close to L3 level as possible;

- In other cases, it will take a long time to realize autonomous driving. It is even more distant to expand from L4 to Geo5 and realize fully autonomous driving L5.

Autonomous driving is now a reality

Autonomous driving application

(1) Autonomous driving falls into typical application areas

At present, the application of autonomous driving is in a stage of flourishing, ranging from passenger cars to commercial vehicles, from travel to logistics, from closed scenes to public roads, from intelligent driving to demonstration operations, etc. Here is a brief summary of the commercial application field, as shown in the following figure:

- In the field of personal travel, there are two main applications: automatic driving and automatic parking. For automatic driving, the development is a process of continuously expanding functions and operating scenarios, represented by Xiaopeng NGP; for automatic parking, the parking lot scenario is simple and is developing towards the L4 level, represented by WM AVP;

- In the field of public transportation, there are two main applications: unmanned taxis and unmanned buses. Currently, they are mainly in the demonstration operation stage on simple urban roads, and there is still a long way to go to achieve commercial operation. The purpose is to promote and train the system. Such as Auto X and Pony.ai Robot Taxi, Yutong "Xiaoyu" unmanned bus, Qingzhou Zhihang "Robot-bus", etc.;

-In the field of park services, the main applications include: unmanned delivery, unmanned retail, unmanned sanitation, etc., represented by Cainiao unmanned logistics, JD delivery robots , and Zhonglian unmanned sanitation;

-In the field of park logistics, the main applications are: unmanned driving in mining area transportation, port container trucks, airport ferries, etc. Most of them are currently in the demonstration operation stage, such as Tage Intelligent Driving Inner Mongolia Mining Area, Xijing Tianjin Port, Yushi Technology, Hong Kong Airport, etc.;

- In the field of long-distance logistics: Tucson is at the forefront of trunk logistics and has already carried out demonstration operations in Arizona, the United States. Due to legal restrictions in China, it is currently understood that only Inche Technology has conducted some operational verification on an open test road in Changsha;

-Other application areas include unmanned agricultural machinery, public security and fire fighting and other special scenarios, XAG's unmanned seed drill, and AiShang Group's unmanned patrol police car.

Typical applications of autonomous driving (commercial)

(2) Autonomous driving reaches the typical application stage

Based on the level of autonomous driving and the level of geo-fencing, I have classified the typical applications of autonomous driving in another way, as shown in the following figure:

- Applications such as unmanned agricultural machinery, unmanned patrol, unmanned delivery, and unmanned sanitation. Due to the simple scene environment and low speed of automatic driving, it is technically possible to achieve unmanned driving. With the reduction of hardware costs, it is expected to be fully commercialized;

-In closed scenarios such as mining areas, ports, airports, and parking lots, autonomous driving is combined with the scenario management system to provide autonomous driving reliability and is expected to be fully commercialized.

- Passenger car autonomous driving, due to the complexity of the scenarios, can only be continuously expanded in terms of functions, and commercial value can be achieved by providing users with intelligent experience;

-Unmanned taxis and buses are unlikely to be commercialized in a short period of time due to the complexity of the scenarios. Demonstration operations are more used for publicity by enterprises and local governments, as well as training of autonomous driving systems.

Typical applications (technologies) of autonomous driving

postscript

The road to autonomous driving is a process in which technological development and commercial interests promote each other, and neither is indispensable. In this process, enterprises, governments, and individuals all play important roles. Let us work together for the better life that autonomous driving will bring!