The large model is accelerating onto the car. The China Academy of Information and Communications Technology previously released the first domestic automotive large model standard, which mainly covers three capability domains. Among them, scene richness focuses on evaluating the support of car models for sub scenarios such as intelligent cabins and autonomous driving, while capability support focuses on the performance of car models in artificial intelligence technology capabilities such as perception, understanding, reasoning, and generation. Application maturity mainly evaluates the application of car models in system ecology, deployment customization, scene adaptation, and other aspects.
There are already many application cases of large models in intelligent cabins and autonomous driving. The large model plays a crucial role in the intelligent cockpit, and its role is reflected in multiple aspects: firstly, it provides a natural and intelligent dialogue experience. Through dialogue generation technology, the large model enables the cockpit system to communicate more effectively with the driver, providing a more natural and intelligent dialogue experience.
This can not only reduce the driver's driving risk, but also enable the driver to obtain the necessary information at any time during the driving process. Secondly, it is driver sentiment analysis and state monitoring. The large model can analyze the driver's speech, facial expressions, behavior and other information, understand the driver's emotional state and attention level. Based on these analyses, the intelligent cockpit can provide appropriate emotional regulation and driver monitoring functions, thereby improving driving safety.
The third is to predict and optimize driving behavior. Through large models and machine learning techniques, the intelligent cockpit can predict the behavior and needs of drivers, and provide appropriate driving advice and auxiliary control based on factors such as driving environment and road conditions. For example, by analyzing the driver's driving habits and vehicle performance, the intelligent cockpit can optimize fuel efficiency and safe driving strategies. The fourth is personalized recommendation and information services. Based on a deep understanding of user profiles through large models, intelligent cabins can provide personalized recommendations for passengers. For example, the intelligent cockpit can adjust seat, air conditioning, music, and other settings according to the preferences of passengers, improving ride comfort. There is also speech recognition and control. Large models improve the accuracy and real-time performance of speech recognition by learning from massive training data. In the intelligent cockpit, speech recognition technology based on large models can achieve seamless integration of vehicle control and information services, making it more convenient for drivers to operate the vehicle during driving.
The role of the large model in automatic driving is reflected in: first, scene perception. The large model can process and analyze huge amounts of data, thus refining valuable information and enhancing the perception accuracy of the auto drive system. These models are capable of interpreting images and video data around vehicles, identifying key elements such as road signs, traffic signals, and other vehicles, thereby ensuring the accuracy of system positioning and path planning capabilities.
The second is path planning and optimization. Large models can analyze and learn the characteristics and traffic conditions of different road networks, including traffic congestion and road conditions. These models are used for real-time route planning and path optimization, ensuring that autonomous vehicles choose the best route and avoid potential traffic problems.
The third is anomaly detection and traffic flow optimization and scheduling: Large models can detect and identify abnormal situations during vehicle operation, and handle them in a timely manner to ensure driving safety. Large models can also analyze and optimize traffic flow, improve traffic efficiency and energy utilization efficiency. By collecting a large amount of traffic data and applying large models, intelligent traffic signal control, intersection optimization, and precise and real-time vehicle scheduling can be achieved. The fourth is prediction and decision support. Large models can use machine learning and deep learning methods to predict different traffic scenarios and provide decision support for autonomous vehicles. For example, by analyzing historical traffic data and real-time sensor data, large models can predict future traffic flow, road conditions, etc., in order to optimize vehicle decision-making and driving strategies.
Many companies are pushing large car models to the market
At present, many enterprises are promoting the application of large models in automobiles. For example, Geely recently demonstrated its self-developed AI digital chassis that can "automatically control cars and avoid risks", becoming the first car company in the industry to have the full system capability of "constructing AI cars with AI frames". Geely's self-developed AI digital chassis showcases the cross domain integration capabilities of self-developed power chassis, including domain control, wire controlled steering, wheel electric drive, intelligent driving, and AI large models. It possesses the highest dimensional intelligent driving and active safety capabilities of digital chassis. Through the perfect integration of the world's first AI large model and digital chassis, Geely Automobile has taken the lead in achieving the world's first autonomous driving drift.
The Geely AI digital chassis has a reaction speed of only 4 milliseconds, which is 25 times faster than the maximum reaction speed of humans. It can achieve "automatic vehicle control and avoidance" under extreme reactions, providing users with "active avoidance and never losing control" safety protection. IFLYTEK has launched the "Flying Fish Scene Intelligent Cockpit System", which is a fully stack self-developed technology that deeply integrates the capabilities of summarizing, reasoning, contextual understanding, and complex content generation of large models. It is closely integrated with various driving scenarios such as driving, communication, and entertainment, and combined with voice and visual interaction, can bring users a more natural, free, and intelligent cabin experience. At present, the iFLYTEK Spark model has been applied in multiple models of well-known automotive companies such as Chery, GAC, and Great Wall. In addition, models such as the Star Journey ES, Haobo GT, and Weipai Blue Mountain are equipped with this technology.
Tencent has previously released a "Global Intelligence" solution for the automotive industry, which mainly covers five core scenarios including automotive research and development, production, marketing, services, and enterprise collaborative office. By introducing core technologies such as cloud computing, AI, and mapping, Tencent aims to promote the automotive industry to achieve "Global Intelligence". In terms of model capabilities, Tencent Cloud's automotive industry big model is based on the self-developed hybrid big model of the entire chain, incorporating massive professional data from the automotive industry for pre training, fine tuning of vertical tasks in the automotive industry, and reinforcement learning. It has performed excellently in Chinese reading comprehension, end-to-end Q&A, and automotive industry related tasks. In terms of specific applications, the "Global Intelligence" solution covers multiple fields. For example, in the field of sales and marketing, large models can quickly generate customer profiles and tags based on past online consultations, conduct lead ratings, and intelligently analyze sales data over a period of time.
In the field of customer service, the intelligent customer service assistant supported by the big model can assist after-sales in providing faster and more professional answers to difficult problems related to car use and maintenance. In the research and development phase, Tencent Cloud AI Code Assistant can help write code, supplement code, diagnose and test code, achieving automatic generation of over 30% of the code in automotive research and development, and improving overall human efficiency by at least 7%.
It can be seen that the big model is comprehensively enabling the automobile industry, including automobile design, production, sales, as well as the improvement of intelligent cockpit, auto drive system capability and other aspects. Of course, while accelerating the installation of large models, they also face some challenges, such as high hardware and computing resource requirements, which will increase the cost of automobile manufacturing. At the same time, as it is a relatively new technology, there is still room for improvement in maturity and reliability. However, I believe that with the efforts of the industrial chain, various problems will be solved.