Research on Application of Artificial Intelligence in Development of Intelligent Driving Engineering Technology
- Chinese keyword
- Artificial Intelligence, Intelligent Driving, Environmental Perception, Decision Planning
- English keyword
This paper briefly introduces the latest frontier progress and background of this problem in the current academic research and scientific and technological development, as well as the key difficulties and challenges it will face in the future.
After more than half a century of development, artificial intelligence technology has profoundly changed the level of scientific and technological development and the way of human life. Especially since 2010, the application of in-depth neural networks has brought artificial intelligence technology into a brand-new stage of development, gradually changing from laboratory to commercial stage, and has developed rapidly in the fields of mobile Internet, intelligent manufacturing, medical big data, supercomputing, sensor networks, brain science, etc. Whether AlphaGo beats the world's Go players or Google's driverless vehicle tests in California, it shows the power of artificial intelligence technology. Artificial intelligence has become the new focus of international competition and the new engine of economic development, bringing new opportunities to China's rapid development.
Driven by the global trend of electrification of automobile power, intelligent control and information networking, intelligent automobiles have become the forefront hot spot in the field of international automobile engineering and the core of future market competition. They are also important measures and development opportunities for China's automobile industry to realize the "Made in China 2025" energy-saving and new energy automobile strategy and supply-side reform. At present, governments, automobile manufacturers, automobile parts manufacturers and Internet manufacturers all over the world are laying out intelligent driving technology. However, due to the complex driving environment that the car needs to face in the driving process, it is difficult to meet the development requirements of automatic driving by relying on traditional environment perception technology and rule-based decision planning control algorithm, and the real fully automatic driving car has not yet appeared.
Applying artificial intelligence algorithm to the development of automatic driving will bring a brand-new development path to automatic driving. Based on the latest research results in the fields of automobile dynamics theory, automobile control, man-machine co-driving behavior analysis and artificial intelligence, such as in-depth learning, information fusion and enhanced learning control, Driven by the functional requirements of the whole vehicle, sensors such as cameras, millimeter wave radars, lidars and ultrasonic radars are selected, and technologies such as deep neural network learning are adopted to realize traffic environment perception and intelligent decision-making planning, thus completing the development of technologies related to automatic driving:
1) Dynamic and static target detection and tracking technology based on depth learning and multi-sensor fusion
This paper studies the cost-effective multi-source sensor autonomous sensing system scheme, establishes an effective target identification and classification method based on deep learning to process unstructured data, and realizes autonomous dynamic and static target detection and tracking algorithms in complex driving environments with Chinese road conditions, such as mixed flow of people and vehicles.
2) Semantic level understanding and local scene generation technology of spatio-temporal multi-dimensional variable scale road traffic environment
Based on autonomous and collaborative environment perception, seamless positioning with high accuracy and behavior information of the vehicle and traffic participants, A method for realize that semantic expression of the road traffic environment around the vehicle, A time and space multi-dimensional variable scale local 3D scene is set up, Characterize and quantify the behavior characteristics of vehicles and their surrounding targets, realize the classification of vehicle driving behavior and target motion behavior characteristics, complete the accurate extraction and effective identification of typical behavior characteristics, and propose a vehicle driving behavior and target motion behavior prediction method based on typical behavior characteristics.
3) anthropomorphic decision-making and cooperative control technology for auto-driving vehicle in complex driving environment
Focus on the key scientific issues of personified automatic driving decision-making and trajectory planning, carry out individual driving and group traffic analysis, and focus on breaking through two key technical issues: (1) autonomous decision-making, trajectory planning and switching of human and vehicle control rights; (2) Vehicle longitudinal and transverse dynamic decoupling and robust control under multi-objective optimization.
4) Safety Evaluation and Test Technology of Autopilot in Virtual and Real Vehicle Environment
According to the safe driving mechanism of self-driving vehicles, a virtual simulation test platform for self-driving and a real vehicle verification test platform are established, and driving safety evaluation mechanisms under different platforms are formulated. Emphasis is placed on breaking through the high-restoration virtual test scenario construction technology and a complete intelligent driving vehicle test evaluation system.
This paper briefly introduces the significant influence and leading role on the scientific and technological development in this field or other related cross-cutting fields after the breakthrough of this problem, as well as the possible major scientific and technological, economic and social benefits.
Because the traditional automatic driving perception and decision-making technology cannot meet the technical needs of the current intelligent driving vehicle development, Combining artificial intelligence technology with automobile industry, Through the deep neural network output multi-target detection results and reduce the target false recognition rate and computational requirements, Through pixel-level target detection, accurate recognition and prediction of small or severely occluded targets are realized, and a multi-dimensional variable scale local scene semantic expression method of people-vehicles-roads in complex environment is completed. Using high-precision static information of driving maps and dynamic estimation of driving behaviors, a path dynamic planning method based on traffic scene reconstruction is designed. The development and application of artificial intelligence technology in the field of automobiles will greatly promote the development of image recognition technology, speech recognition technology, multi-sensor sensing fusion technology, enhanced learning technology and big data analysis and mining.
The combination of intelligent automobile manufacturing and artificial intelligence technology will drive the development of the entire automobile industry, including sensor manufacturing products, chip manufacturing industry, wireless communication industry and Internet industry, and will form a 100 billion-level industrial group in the future. The emergence of fully automatic driverless vehicles will break the current social travel mode, reshape the road planning pattern, promote the consumption mode of shared travel, achieve the goal of zero casualties and zero accidents in future traffic, and make the future travel mode safer, more convenient and more environmentally friendly.