Heavy perception and light map is becoming the choice of more and more car companies. Xiaopeng announced that by the end of this year, it is expected that there will be 50 cities to achieve the landing of XNGP, smart self has also announced the IMAD intelligent driving technology planning, will soon open "go to high-precision maps NOA" public testing. However, there is also the opposite, in early August, Guangzhou Automobile Group invested 40 million yuan layout of high-precision maps. Why do enterprises have very different choices for perception and map? Perception and map will show what kind of development?
Heavy perception is on the rise
Heavy perception is on the rise, and NOA (Pilot Assisted Driving System) is highly favoured by car companies, and different system names have been derived from it, such as Peng's NGP, Azure's NOP, Huawei's NCA, Great Wall's NOH, and Ideal and Tesla's directly named NOA, but essentially it is basically the same, heavy on perception and light on maps. With more and more enterprises starting NOA public testing, this year is also known as the "NOA year". At present, there are enterprises including Xiaopeng, Huawei, millimetre wise, ideal and other enterprises have opened the city NOA test. Recently, the smart self announced IMAD intelligent driving technology planning, according to the plan, in September, the smart self will start to go to high-precision maps NOA public testing; in October to start the city NOA public testing; in 2024, the commuting mode of a hundred cities open; in 2025, the smart self IM AD into the era of Door to Door (the whole scene of the commute) for most of the scenarios. 
Xiaopeng Automobile also announced recently that by the end of this year, 50 cities are expected to realise the XNGP landing. In March this year, Xiaopeng G9 and P7i Max version opened point-to-point city NGP in Shanghai, Shenzhen and Guangzhou, and Xiaopeng P5P series also opened point-to-point city NGP in Shanghai's high-precision map coverage area. he Xiaopeng, founder and CEO of Xiaopeng Automobile, emphasised that the mapless XNGP is undergoing large-scale testing nationwide, and that the XNGP will surely redefine the smart driving. He Xiaopeng revealed that the city scope of XNGP will be expanded from a few cities now to dozens of cities within the year, and next year the goal is to increase to most of the cities in China. Xiaopeng will also launch the "AI chauffeur" mode to all XNGP users in the fourth quarter of this year. This means that in areas not covered by high-precision maps, Xiaopeng Auto can personalise intelligent driving routes according to user habits, and realise high-level assisted driving from point A to point B. At this year's Ideal Family Technology Day, Ideal Auto announced that it will test the city NOA function in Beijing and Shanghai. It is understood that Ideal AD Max 3.0 gets rid of the dependence on high-precision maps through large-model AI algorithms, so that the vehicle's automatic driving system can sense, make decisions, and plan in real time like a human driver. Earlier this year, Great Wall Motor's millimetre-end Zhixing announced that the city NOH function will land in the third quarter of this year, equipped in Wei brand Moka DHT-PHEV and Blue Mountain in turn. Huawei also released the HUAWEI ADS 2.0 system in April, and also said that the city NOH has landed in Shenzhen, Shanghai and Guangzhou. Yu Chengdong, Huawei's Managing Director, CEO of Terminal BG and CEO of Intelligent Vehicle Solutions BU, revealed that urban NCA, which does not rely on high-precision maps, will land in 15 cities in the third quarter of this year, and that it will increase to 45 cities by the fourth quarter. Baidu Apollo launched the city smart driving product Apollo City Driving Max, is expected to be equipped with the product of the relevant models will be delivered this year. Heavy perception is more in line with current development needs The debate of heavy perception and light map has been going on for some time. As early as 2019, Tesla founder and CEO Musk publicly "blasted" traditional high-precision maps and steered Tesla's Autopilot solution towards a purely visual perception solution. Musk believes that high-precision maps are a "very bad" idea. When the road changes a little bit, high-precision maps become obsolete. Last year, Huawei and Xiaopeng also said they wanted to get rid of their reliance on high-precision maps. At the same time, too much reliance on high-precision maps and vehicle-road coordination will limit the ability to improve the automatic driving and intelligent driving, which also makes the car companies generally choose to re-perception of this route. In the view of industry insiders, the update and maintenance costs of high-precision maps are high, and there is a certain lag, which is unable to adapt to changes in the road in a timely manner. At the same time, vehicle-road collaboration requires real-time communication between vehicles and road infrastructure, but in reality the popularity of infrastructure varies, and vehicles cannot always rely on communication. Based on this, over-reliance on high-precision maps and vehicle-road collaboration can bring about limitations in automated driving systems, preventing vehicles from adapting to a variety of complex scenarios and road environments, and limiting the development of automated driving technology. 
The higher cost of high-precision maps is another commercial thinking of a group of car companies that choose to re-sense. "One high precision, high investment everywhere." From a commercial perspective, relying heavily on high-precision maps requires more investment, and at this stage, bypassing high-precision maps is more in line with commercial logic for car companies. Industry insiders point out that high-precision maps require a lot of mapping work and update and maintenance costs. The White Paper on High Precision Maps for Intelligent Networked Vehicles shows that, using traditional mapping methods, the mapping efficiency of a decimetre-level map is about 500 kilometres per vehicle per day, with a cost of about $10 per kilometre, while the mapping efficiency of a centimetre-level map is about 100 kilometres per vehicle per day, with a cost of up to $1,000 per kilometre, or $100,000 per day. In urban scenarios, roads change frequently, and the accuracy of high-precision maps cannot meet the actual demand. Yu Chengdong said: "The cost of high-precision maps is very high, just in Shanghai, the collection of one or two years, 9,000 kilometres, are not completely covered Shanghai, the cost of national coverage is very high." In addition, car companies using high-precision maps need to pay the mapmaker authorisation fee or service fee. It is understood that the authorisation fee ranges from about 200 to 1,000 yuan per car per year; the service fee ranges from about 100 to 500 yuan per car per year. This is undoubtedly a high cost for car companies. More critically, the perception system can not only support the landing of automatic driving technology, or an important part of the intelligent cockpit, and the intelligent cockpit is now very high market acceptance, and even to a certain extent become an important pivot point of the Chinese auto brands to compete for the market and enhance the competitiveness of the brand, and, therefore, more Chinese car companies will choose to strengthen the layout of the perception system. Perception and maps need to work together
Under the trend of heavy perception, high-precision map is really no one? This is not the case, there are still enterprises increasing investment in high-precision maps. in early August, Guangzhou Automobile Group invested 40 million yuan in the layout of high-precision maps, with the aim of further improving the industrial chain layout of the core areas of the intelligent network connection, intelligent driving and high-precision maps. "Perception and map are complementary, perception and map should exist at the same time, rather than saying that there is no perception if there is a map, or that the map should be abandoned if there is perception." Jiang Rui, general manager of Auto Business Centre of Gaode Maps, believes that perception solves the near-field problem, i.e., the problem around the car or within more than 100 metres or 200 metres in front of the car, while the map has to solve the global problem, and the map gives all the prior a priori information, which is a great complement to the near-field perception. "There's no way for a car to know what's happening five kilometres away if it's just relying on sensors. So it's not that the market doesn't need a map, but it needs a lighter, live map. For high-level autonomous driving to get better, it needs the coupling of perception and maps, both working together. "The perception system represented by the camera mainly solves the problem of timeliness, while the high-precision map mainly focuses on directionality. When autonomous driving develops to a high-level stage, the two must work together to present better results, not an either/or relationship." For example, from point A to point B, the vehicle not only needs a clear driving route, but also needs to understand the road conditions in real time, so as to quickly reach, "need to play a role in the eyes and ears." At the same time, to better achieve high-level automatic driving, but also need to use the power of the network (cloud), to provide remote computing and other help. "The perception system can only help vehicles achieve simple assisted driving functions, high-level automatic driving can not bypass the support of high-precision maps.
