Shared todayAI seriesIn-depth research report: " Vehicle-mounted holographic digital human: AI+Agent New scenes, new directions for holographic projection》.
(Report produced by: Guosheng Securities)
Total report: 13 pages
The rapid progress of AGI has always been the focus of market attention. It has maintained high market popularity this year. From computing power, games, marketing to medical care, education, etc., the market has been looking forward to new application breakthroughs. We have observed that as sales of new car-making forces increase month by month; Qualcomm releases the latest Snapdragon 8 Gen 3 and Snapdragon Become an important part of the three-screen integration of "one person, one car and one family"
1. Vehicle-mounted holographic digital human: AI Agent new scenarios, new human-vehicle interaction experience
The lifelike digital human in the smart cockpit is expected to bring a "human-human interaction" experience to drivers. The NOMI robot previously installed in Weilai cars once became the main selling point of the Weilai brand. NOMI's ever-changing expressions and various decorative peripherals made the NOMI robot successfully "out of the circle". We believe that a smart cockpit digital person is defined by being highly friendly, having a high enough level of intelligence, and being able to help drivers complete simple (such as answering phone calls, playing music) or complex (such as navigation, life assistant, car maintenance) tasks. Core takeaways.
In the future, human-computer interaction will not stop at cold screens and speakers. Traditional human-vehicle interaction uses the central control screen as the medium. NIO's NOMI materializes the intelligent assistant into a robot image. If it is further embodied, digital human is undoubtedly the best route. Take Prism Hologram's digital human interaction platform as an example. It is based on cloud and local dynamic behavior databases, drives virtual human images according to different scenarios, and expands the use scenarios and value of virtual humans. At the same time, it provides a large number of peripheral materials based on digital assets, supports personalized image selection, dress matching, etc. of digital virtual people, and creates a user-specific intelligent virtual companion.
Why are digital people in smart cockpits being catalyzed now? We believe that the explosion of generative AI gives digital people in smart cockpits a "soul", holographic HUD and high-computing intelligent modules add "flesh and bones" to digital people, and the cockpit is intelligent In this era, digital people are given more space to display their abilities.
1.1 AIGC: the “finishing touch” of digital people
With the development of AIGC generation algorithms and natural language models in the past two years, AIGC has been widely used in the generation of text, images, audios, games and codes. The output content has a high degree of completion and can be widely used in commercialization middle. At present, the mainstream 2D simulated digital people are mainly presented through pictures or photos after completing the holster, superimposing mouth shapes and a few movements, and then combining emotional analysis and decision-making mechanisms. In this process, AIGC technology has the following help for the generation of digital people. :
⚫ Improves the efficiency and convenience of content creation, and reduces costs and thresholds.
⚫ It enriches the diversity and personalization of content creation and meets the different needs and preferences of users.
At present, the various personalities and behaviors of digital people are all scripts edited by the team, and the digital people that users see in front of the stage are only the output of the operations team behind them. For example, many digital UP hosts on Hualihuali have real people dubbing and interacting behind the "holsters". At the beginning of Double Eleven last year, we also noticed that many Taobao brand flagship stores used digital anchors to promote goods. Although these digital human anchors are synthesized by AI and have no sense of inconsistency in their movements and mouth shapes, and can achieve 24/7 live broadcast and intelligently generate broadcast video stream content, the interactive experience is still relatively rudimentary.
In November 2022, ChatGPT quickly became a "popular hot chicken" in the AI industry as soon as it was launched. In February, the number of users exceeded 100 million, making it the second fastest growing application in history. This is mainly due to its powerful language generation capabilities and diverse application scenarios. ChatGPT can not only answer various questions, but can also write scripts, novels, instructions, and even correct code errors. In the process, it has shown a very high "emotional intelligence" and can conduct multiple rounds of conversations and reply to messages. Smooth and smooth.
On March 2, 2023, OpenAI announced the opening of the API interfaces of ChatGPT and whisper. ChatGPT API is a language model interface for developers. It allows developers to use the GPT 3.5 Turbo model to generate natural dialogue and text. The advantages of ChatGPT API are that it has continuous model improvement, low price, and optional dedicated capacity.
Whisper API is a speech recognition interface for developers that allows users to convert speech to text. It uses a deep neural network model trained on large amounts of speech data to achieve high accuracy. Whisper API can recognize audio files in many formats, including mp3, mp4, mpeg, mpga, m4a, wav or webm. The advantages of the Whisper API are its extremely high speed, easy on-demand access, and reasonable price. This move will make the interactive experience of digital people a big step forward.
Vertical city models are penetrating into cars. On July 31 this year, Geely Automobile Group revealed at an internal business conference that Geely will release "the industry's first full-stack self-developed full-scenario AI model" in the second half of the year. The AI smart electric hybrid family sedan "Geely Galaxy L6" was also launched in September. The AI out-of-car voice interaction technology equipped on the Galaxy L6 can support the opening and closing of windows, air conditioners, and trunks through voice commands within a distance of 1.5 meters outside the car, as well as customizing interesting voice packages and lexicon, creating a seamless experience inside and outside the car. A new social scenario for communication barriers. The new car will also launch Geely map navigation, which supports traffic light countdowns. Even the fragrance system in the car can be intelligently identified and released according to different scenarios. We believe that in the increasingly fierce market competition, AI will be the key to car manufacturers upgrading their intelligence and making products. The best choice to create differentiation, automotive AI applications are expected to explode in 2024.
1.2 HUD hardware and edge computing hardware are mature, and digital humans have “flesh and blood”
The development history of automobile cockpit hardware is the epitome of the development of consumer electronics products. From hardware-driven to application-driven, from local basic entertainment and navigation to intelligent interconnection of people, vehicles and objects, we divide the automobile cockpit into four development stages.
1) The button era. In 1924, Chevrolet produced the first car equipped with a radio. In 1941, Cadillac launched the 75 Series equipped with electric windows. In the 1960s, Ford and Motorola "introduced" tape players into cars. ;
2) Screen era. At the beginning of this century, with the application of LCD screens, displayable content became an important step in the development of cockpits: then software applications began to develop vigorously, and the central control car computer that supports Kaili De navigation became a popular modification. Solution; Whether to support Apple Carplay and Baidu Carlife is an important consideration factor for consumers when purchasing a car model;
3) The era of screen interconnection. Car-to-machine applications have further developed, the concept of Internet of Vehicles (V2X) has emerged, autonomous driving and smart cockpits have begun to merge, and independent car-machine systems that rely on the computing power of on-board chips have emerged;
4) The era of large and multiple screens. Generative AI further empowers human-vehicle interaction and lights up the soul of the car-machine; more screens and larger sizes further increase the density of human-vehicle interaction; L2 and L3 high-end autonomous driving Solutions are becoming increasingly mature, autonomous driving and smart cockpits are deeply integrated, and the concept of "smart driving" is getting closer.
The interaction potential of two-dimensional touch screens is being fully exploited, and product designers are beginning to consider three-dimensional human-vehicle interaction. Two important components must be mentioned here:
1) Head-up display (HeadUpDisplay, HUD). HUD was originally used on fighter jets to improve pilots’ driving efficiency. Pilots can obtain flight data without frequently looking down at instruments, and it also facilitates pilots’ aiming operations. This product logic has been translated to cars. It has become a new tool to replace the traditional instrument panel. The driver does not have to lower his head and can check the vehicle speed and navigation information from the front of the field of vision, which greatly improves the driving experience.
The information generated by the holographic aerial display (AID) through the image unit (digital people, driving information, navigation, etc.) is projected in front of the driver through the windshield (or not through the windshield) by the off-axis reflection system. AID uses microstructured light field reconstruction technology to discretize objects into high-resolution spatial light spot arrays. Through light diffraction propagation, interference superposition and complex amplitude control of the light field, the discrete light spot arrays are losslessly reconstructed in the In the air, high-fidelity images can be observed without any scattering from the medium, thus achieving medium-free imaging. The maturity of AID technology has provided excellent conditions for three-dimensional interaction between digital people.
2) Edge computing power module. The development of the AI flywheel is moving forward. From GPT-3.5 surprising the world at the beginning of this year, to OpenAI officially opening the plug-in function on May 14, we have witnessed the rapid iteration and evolution of large models in the cloud. At this point, AI large models are already in a critical iteration period of rapid evolution from “toys” to “tools”. How to let large models penetrate into various vertical scenarios, how to use large models at a lower cost, and how to bring more scenarios and users into contact with AI have become the next focus of development. When AI is implemented in actual scenarios, the importance of edge computing is accelerating. Edge computing has natural advantages in cost, latency, and privacy. It can also be used as a bridge to preprocess massive and complex requirements and guide them to large models. The status of edge computing as a blood vessel for AI involving thousands of scenarios is accelerating and clear. In the past, edge-side devices mainly deployed communication capabilities. Now, driven by edge giants represented by Qualcomm, edge infrastructure such as computing power and AI capabilities are being accelerated. At the same time, the computing power networks and edge resource pools of the three major operators The system also laid the prototype for the gradient distribution of computing power in our country. Looking to the future, edge computing power will begin with the increased demand brought about by AI. It will also empower applications, connect more users, and accelerate AI development and iteration.
1.3 The higher the level of autonomous driving, the more important the human-vehicle interaction becomes.
The classification of autonomous driving levels is currently widely accepted at home and abroad by the SAE (Society of Automotive Engineers) classification, which has a total of 6 levels from Level-0 to Level-5, with Level-0 being the lowest level and Level-5 being the highest level. As shown in the figure below, from no automation, to "freeing feet", "freeing hands", "freeing eyes", "freeing the brain", and finally reaching complete "driverless".
We believe that the more advanced autonomous driving is, the more frequent, stable, and efficient human-vehicle interaction is required. In the LO-L2 era, since autonomous driving can only complete some basic driving tasks, such as lane keeping and adaptive cruise, drivers need to pay attention to traffic conditions at all times and do not have much energy to pay attention. For human-vehicle interaction, the smart cockpit only needs to provide the driver with basic navigation, vehicle monitoring, and entertainment interactions. In the era of high-level autonomous driving, the driver does not need to focus on the driving itself at all times. Autonomous driving can complete most of the driving tasks in regular scenarios. At this time, the smart cockpit needs to shoulder an important task-to keep close contact with the driver. Contact, so that when autonomous driving cannot complete the driving work, the driver can intervene in the steering wheel in time to avoid accidents, and Anthropomorphic, three-dimensional, highly intelligent HUD and even holographic digital people are such Helpful interaction method in the scene. Compared with the artificial intelligence on the screen, the holographic digital human is more friendly and agile, supports customized appearance, and can attract young users: Compared with the NIO NOMI holographic digital human that is independent from the screen, it is more The sense of technology has the opportunity to open up a new market space among young user groups.
2.AI Agent empowers human-vehicle interaction in the next era one by one
AIAgent (artificial intelligence agent) has entered a period of accelerated growth driven by the rapid development of large models. LLM applications such as GPT still have limitations when updating information, processing multiple rounds of dialogue, and facing complex tasks. AI Agent overcomes these limitations by adding three major capabilities: planning, memory, and tool use, greatly expanding the application of large language models. scope, making it capable of more complex tasks, which provides the possibility for AI Agents to act as "assistants" in human daily life.
Cars are one of the most potential application scenarios for AI Agents. When driving, car owners need to operate the car quickly and accurately. In the era when the cockpit was not yet intelligent, such operations relied on mechanical buttons and operating levers. In the future, human-car interaction will increasingly focus on central control screens. Traditional operating habits need to change. The capabilities of AI Agent meet this need: quick response, accurate understanding of commands and hands-free operation. The intelligence given by AI can further extend the capabilities of Agent. These abstract capabilities can be embodied in application scenarios and can be AIAgent's capabilities are summarized into five levels: auxiliary operation, body data collection & visualization, car services (navigation, maintenance, insurance, etc.), life assistant, and embodied intelligence.
The penetration rate of smart cockpits and HUDs is increasing rapidly, and the market prospects of digital humans in smart cockpits are optimistic. NIO NOMI is quite popular. According to the 2022 NIO second-hand car data compiled by Zhijian Cheyan, the proportion of NOMI optional equipment has reached more than 80%, which is the same as augmented reality and second only to NIOPilot autonomous driving. Feature pack. We have always emphasized that only when the flywheel of application-computing power (that is, revenue-cost) is rotated can generative AI develop healthily. Among consumer products such as cars, since the price of the car itself is relatively high, the marginal cost of several thousand yuan for this type of smart assistant is relatively low, so it can attract more users to choose it.
The subscription system for vehicle functions that has emerged in recent years has opened up new profit ideas for AI products. For example, Huawei ADS 1.0 can be purchased for a one-time purchase of 32,000 yuan, the annual subscription is 6,400 yuan, and the monthly subscription is 640 yuan; ADS 2.0 can be purchased for a one-time purchase of 36,000 yuan, and the annual subscription is 7,200 yuan. Yuan, monthly subscription is 720 Yuan. We believe that the in-vehicle function subscription system is expected to be an important business model for smart car consumption in the future. Based on this payment model and taking into account the penetration rate of smart cockpits, we can estimate the market space for domestic smart cockpit digital people.
⚫ Estimation logic one: Passenger car market growth. According to data from the Ministry of Public Security, China's car ownership will exceed 330 million vehicles in September 2023, and may exceed 400 million vehicles in 2025; according to data forecasts from Guanyan Report Network, my country's passenger car sales from 2023 to 2025 will be 2389/2437/2485 Thousands of vehicles.
⚫ Estimation logic two: smart cockpit penetration rate. Since the proportion of rear-mounted HUDs is very small, we only consider the front-mounted market here. According to IHS Markit data, the penetration rate of smart cockpits in domestic new cars will be 60% in 2022. Since then, as the penetration rate of new energy vehicles has increased, driving fuel vehicles towards intelligence, the penetration rate of smart cockpits is expected to further increase to 66%/72 in 2023-2025. %/76%.
⚫ Estimation logic three: HUD penetration rate. According to IHS Markit, the HUD penetration rate will only be 10% in 2022. It was previously monopolized by overseas suppliers. Market players have emerged this year, and combined with the decline in CR5, the HUD penetration rate is expected to increase rapidly.
⚫ Estimation logic four: smart cockpit digital human pricing. Considering that the smart cockpit digital person is not a necessity for driving, but an optional consumption for upgrading the driving experience. At the same time, considering the high research and development costs behind the smart cockpit digital person, we believe that the monthly subscription price should be set at 40-60 yuan/each Monthly or 400-600 yuan/year is more appropriate.
⚫ Estimation logic five: smart cockpit digital human penetration rate. We can calculate penetration rates of 20%/50%/80% based on pessimism/neutral/optimism respectively.
Total report: 13 pages