The year 2023, which is about to end, is an extremely "involution" year for the field of automotive intelligence. Driven by the involution of Huawei, Xiaopeng and other car companies, the installation rate of intelligent driving in China generally exceeds 50%, and the penetration rate of intelligent cockpit standard equipment reaches 9213%。It can be said that 2023 is the year of the popularization of automotive intelligence, and low-level intelligent driving and non-intelligent cockpit have become a minority product in the market. However, after the popularization of intelligence, the next step of car companies will be to focus on technology iteration, so that consumers can harvest more competitive and cost-effective products.
From the perspective of the two main aspects of intelligence (intelligent driving and intelligent cockpit), 2024 is likely to be an era of main experience. That is, the high-end intelligent driving function represented by NOA in urban areas will be transferred to the medium computing platform, and the ultimate cost-effective solution will reduce the cost of getting on the carThe intelligent cockpit is beginning to connect to multiple scenarios. In the route of technology iteration, what will be the trend next year?
After entering the stage of urban NOA popularization, a strange phenomenon has emerged among domestic car companies, and car companies that achieve similar urban NOA functions will choose stronger computing hardware to achieve it. Huawei ADS 2The MDC 610 used by the 0 system has a computing power of 200 tops, which supports the NCA function in urban areas. Xpeng has the same urban NOA function, and it uses 2 NVIDIA Orin-X chips, with a total computing power of 504TOPS. NIO even uses 4 Orin-X to achieve intelligent driving, with a total computing power of 1000+ TOPS. With the same user experience, the computing power between different solutions can reach more than 5 times, which is obviously illogical.
At this stage, a problem facing chip and solution providers is that the computing power cannot be improved to a certain stage. At present, the problems faced by urban NOA are not limited to computing power, but also lie in algorithm architecture and hardware configuration.
We can take a look at the hardware configuration that now supports NOA in urban areas, and the mainstream solution is the combination of computing power chips (Orin-X) and LiDAR with more than 200TOPSThe intelligent driving solution that supports high-speed NOA uses 84-128TOPS computing power chip (Orin-N or Horizon Journey 5) + non-lidar combination. It can be seen that the NOA scheme in mainstream urban areas adopts radar perception assistance + enhanced computing power mode.
From the perspective of hardware, the demand for computing power of NOA in urban areas is concentrated in complex roads in urban areas, resulting in the escalation of the complexity of perception data, and the chip needs more computing power to process complex scene elements, including identifying obstacles. From the perspective of software, the mainstream intelligent driving scheme at this stage uses the BEV+ obstacle recognition network, and the core point of the calculation is to identify the type of obstacle and mark the priority of obstacles.
Horizon and NVIDIA are both promoting the end-to-end algorithm framework of BEV+Transformer, which identifies road associations and the trajectories of road participants through the occupancy network**, and the structure of the two algorithm frameworks is different. BEV+Transformer takes advantage of the perspective effect of vision and carries out ** in the BEV space, which changes the traditional perception algorithm.
At present, the mainstream perception algorithm focuses on the camera to perceive the 2D picture, the lidar to identify the position, and the chip to integrate the 2D and 3D sets. bev+transformer can reduce the errors of recognition, perception and ** in perception tasks, achieve end-to-end optimization through neural networks, and promote rapid iteration of perception models through self-learning and Xi.
Based on this new algorithm framework, the number of automotive cameras will be increased, and the role of radar will be reduced, thereby reducing the computing power pressure encountered by the chip. At the SoC level, the only requirement for chips is the need to increase cache and bandwidth, which is much easier for chip manufacturers to adapt and fine-tune the existing chip architecture and add underlying software optimization, which is much easier than increasing computing power.
Based on this architecture model, Horizon has developed an intelligent driving solution realized by the division of labor of 4 journey 5 chips, which can realize high-speed urban NOA, which also proves that this architecture can effectively reduce the demand for chip computing power.
In April 2023, DJI launched the L2+ intelligent driving function with a similar strong visual architecture, which can achieve urban pilot driving with 80TOPS computing power. The cost of this scheme is only between 5,000-15,000 yuan, which makes it possible to board the computing power platform.
In November this year, the second-generation HBapilot no-map NOH solution was released at the end of this year, which only needs 72 100TOPS computing power to achieve urban NOH. These new solutions will reduce the minimum computing power level of urban NOA from 200 TOPS to 72-100 TOPS for medium computing power platforms.
It is foreseeable that next year, the urban NOA function will cover the medium and low computing power platforms with the change of the architecture model, and promote the improvement of the loading rate of high-end intelligent driving. Previously, Yu Chengdong said that the new M7 urban intelligent driving package selection rate is 75%, and next year, driven by the new architecture, the urban NOA selection rate will most likely be increased to 90% of the level, of which the main increase range is in the low and medium computing platform models, that is, low-priced models with a price of 10-200,000.
Regardless of the HarmonyOS cockpit system and the Meizu Flyme Auto system, the main evolution point of the intelligent cockpit system at this stage is to break the island effect of bicycle intelligence and focus on the seamless connection between devices. The super desktop and navigation destination perception functions of the Hongmeng intelligent cockpit system use the computing power of the mobile phone and the car machine to connect and use the mobile phone computing power to make car machine applications.
The surging os, HarmonyOS 4 released this year0 is on the basis of strengthening the computing power connection of mobile phones, and opening up more terminal devices. A typical function is the linkage between the car and the whole house intelligence announced by Huawei at the Zhijie S7 press conference, and the car machine controls the air conditioner, audio, TV, lighting and other smart devices in the home through the smart home central control. This form realizes the connection between the Internet of Vehicles and the Internet of Things, and will continue to deepen in 2024.
Next year, the intelligent cockpit service will strengthen the perception of members in the cockpit, and Huawei recently applied for a patent to provide scenario-based services based on the intimacy of the car owner. The patent senses the occupant's pick-up and drop-off location information, ride location information, and communication frequency information with the car owner through the cockpit, so as to convert the corresponding scenario-based cockpit service. This function significantly strengthens the degree of connection of the intelligent cockpit in multiple scenarios, and it is likely to be loaded in the future Hongmeng cockpit system.
Desay Xesh also proposed a new intelligent cockpit system logic, based on the STR energy-saving method, by obtaining the environmental prediction system to determine whether the system is at the right temperature, by turning on the air conditioning to adjust the ambient temperature inside and outside the cabin, the patent also optimizes the temperature performance of the air conditioning in different temperature scenarios.
All in all, the smart cockpit in 2024 will highlight multi-scene connection, connecting more homes, offices and other scenarios through smart home protocols. And the scene optimization in the cockpit provides a more contextual intelligent cockpit experience.