2024 can be described as the outbreak year of urban NOA, Xpeng and Huawei have gradually completed the functional development of multiple cities by the end of 2023, and other car companies are also eager to try in 2024, and the battle is not fierce.
In January '24, Zhiji Auto also officially launched the world's first trial of urban NOA, and users can already experience the amazing performance of urban NOA in depth on the "road of flowers" in Shanghai. Let's take a look, what are the implementation difficulties? And what black technology can solve these problems.
Standing at this juncture in 2024, high-speed NOA is no longer a hot spot of competitionCompared with the high-speed working conditions with flat and stable structure and single structure, the complexity faced by urban NOA is far from an order of magnitude, you may need to deal with:
1.The signal lights of different specifications in different cities still have a bit of "literary atmosphere" from time to time
2.Uncles and aunts who ignore all laws on the road
3.Overly strange urban intersection design
Wait a minute.
Although from the perspective of functional performance, the NOA of urban working conditions is only a simple expansion of the range of high-speed NOA functions, but in fact, from a technical point of view, thisIn fact, it is a process of quantitative change leading to qualitative change, and it is a new upgrade of the current intelligent driving technology stack
High-speed service. 
Urban conditions. 
In the past, the development of intelligent driving systems mostly used rule algorithms, but with the increase in the complexity of the scene, the amount of development and testing that needs to be faced is also increasing day by day. Engineers have their own limits, both physically and psychologicallyThis limit is largely between high-speed NOA and urban NOA.
Therefore, in the process of expanding from high-speed to urban areas, we need a new paradigm that uses AI models and data-driven development methods to further respond to the more complex external environment. Therefore, one of the key points of the urban NOA competition is actually the proportion of the AI model in the entire function. The higher the proportion, the easier it will be to cope with the complex working conditions of the city or expand the scope of urban services.
The first thing that urban NOA has to face is a large number of close-range special-shaped obstacles, which used to be used in high-speed conditions3D rectangular boxTo characterize the surrounding obstacles, the structure is simple, the training is convenient, and the reliability is high. However, in the crowded state of the city, this expression has many drawbacks, which cannot effectively describe special-shaped obstacles, and there may be a risk of collision at close range.
As a result, a large number of them began to be introduced in the industryAI model-aware algorithm for "occupying the network".After entering the laser and camera data of the surrounding location, the surrounding objects can be described in a raster similar to "Minecraft", so as to enhance the refined description of the surrounding environment. Launched by ZhijiDDOD obstacle perception scheme, which aims to output 3D rectangular boxes and raster descriptions at the same time, to have a more hierarchical and detailed depiction of the environment.
In the past, high-precision map schemes were generally used to solve the information of road regulation in urban NOA, but the cost of engineering maintenance and update of this scheme was extremely high, which was not conducive to function expansion.
So now more is starting in the industryPursue the noa solution of the city without a map, and use visual scenes** to replace the role of the map。Similar to the old driver's driving, the location of blind spots such as intersections is scene** to replace the role of the original map and ensure driving safety. Zhiji has also launched an AI model solution for DDLD to achieve this goal, and based on this, it will launch the update of the no-map city NOA in June to accelerate the pace of urban NOA.
The last thing to be solved in the city NOA is:A complex decision-making problemIn the past, engineers generally used rule algorithms to deal with different driving environments, but urban working conditions are too complex, and if rule algorithms are used at the end of development, it is often a dead sheep, so it is difficult to play a sustainable role in urban NOA.
As a result, Zhiji has launched a data-driven planning approach (Dl.p.), to learn the driver's driving behavior, and use the AI model to generate a decision-making trajectory, so as to better cope with special decision-making scenarios.
ddod+ddld+d.l.p."YesLeverage a data-driven approach instead of rule-based algorithmsto achieve compatibility with more complex urban conditions. More importantly, it can make the intelligent driving system not be dragged down by a large number of long-tail scenario adaptation, give full play to the advantages of AI model generalization, and let more urban users experience the urban NOA function with a lower and faster release speed.
Of course, users still need to be cautious enough in the process of using the city NOA function. Any intelligent driving system has its own intelligent driving level of its own products, and all current intelligent driving products, including the urban NOA function, are L2Level 5 products, where users still need to monitor the environment and be ready to take over immediately.
But the good news is that Zhiji has also begun to fully deploy L3 intelligent driving products.
In December 2023, Zhiji Auto's vehicles equipped with Level 3 autonomous driving functions officially obtained a high-speed autonomous driving test license in Shanghai. At the same time, it is also actively applying for the L3 announcement of the Ministry of Industry and Information Technology for the access pilot, and is expected to become the first batch of models to enter the L3 autonomous driving access pilot. L3 related products are also in full swing research and development. When the product level reaches L3, users can further save the energy consumed in the process of monitoring the driving environment, and users will have a better intelligent driving experience.
I believe that this day will not be too far away!