On June 21, 2023, the first Didi Industry-University-Research Cooperation Forum and Gaia Lighthouse Project 2021-2022 Outstanding Project Award Ceremony were successfully held in Beijing. Didi CTO and CEO of Didi Autonomous Driving, Honorary Chairman of Didi Industry-University-Research Cooperation Committee Zhang Bo, Head of Didi Map Business Department and Chairman of Didi Industry-University-Research Cooperation Committee Chai Hua, Director of Didi Technology Ecology and Development Department Wu Guobin, etc. Members of the Industry-University-Research Cooperation Committee attended the meeting. Professor Cai Ruichu from Guangdong University of Technology, Associate Professor Gao Ruipeng from Beijing Jiaotong University, Professor Li Shengbo from Tsinghua University, and Professor Wang Longbiao from Tianjin University were invited to participate in the forum.
Didi's industry-university-research cooperation began in 2016. It aims to discover and define problems with the academic community through the organic combination of cutting-edge academic research and actual business scenarios, and to cooperate and win-win to solve field problems. It aims to promote cutting-edge technological innovation and solve industry challenges. The goal is to realize the transformation of scientific research innovation into industrial value. The Gaia Lighthouse Project is an important project in Didi's industry-university-research cooperation. Through the mechanism of selection, display and incentives, it promotes the transformation and value dissemination of scientific research cooperation results. During the forum, the "Gaia Lighthouse Project 2021-2022 Outstanding Project Award Ceremony" was held to present awards to the five award-winning outstanding project cooperating teachers and colleagues, and took this opportunity to thank the project team for their efforts.
Award-winning projects
(Sorted by the first letter of the project name pinyin )
Environment Perception and Location Tracking in GPS-missing Scenes
Project team: Associate Professor Gao Ruipeng of Beijing Jiaotong University and Didi Maps team
Project introduction: With the development of technology, drivers basically get rid of paper maps and human memory when driving. Technology has greatly facilitated our lives. But at the same time, inaccurate positioning is like a hidden assassin, stabbing suddenly, making the driver feel how inconvenient it will be when inaccurate positioning occurs. This project is based on a large number of real trajectory training. We deployed a deep vehicle position estimation model on the terminal to solve the problem of inaccurate positioning in weak signal scenarios. In the long tunnel scene, we compared the effect with competing products, surpassing competing products by 17%. It has been fully launched, and has jointly produced 9 academic papers (5 of which are SCI), further expanding Didi's academic influence.
Estimation of long-term effects based on short-term surrogates
Project team: Professor Cai Ruichu of Guangdong University of Technology and Didi's car-hailing team
Project introduction: As an important starting point in the operation of online car-hailing, pricing strategy not only affects short-term supply and demand, but also affects the long-term retention of passengers and drivers, which in turn has an important impact on the long-term operation of the platform. Therefore, how to accurately quantify and evaluate the impact of pricing strategies has very important application value for the platform. Starting from the specific business of price strategy, this project developed a new model, which can dig out more effective representations from massive short-term observable factors, and effectively help the evaluation of long-term strategy effects. In addition, the model also fully considers the differences in data distribution between different cities, and uses domain adaptive technology to improve the stability of the model's effect in different cities and break through city restrictions. The research results of the project, compared with the current optimal algorithm, have achieved a relative improvement of more than 50% in the prediction accuracy of the long-term effect of the strategy, and achieved a relative improvement of more than 70% in the accuracy of the urban migration problem, which is the further optimization of the strategy and adjust to indicate the direction.
High-precision positioning of two-wheeled vehicles
Project team: Professor Liu Hui of Wuhan University and Didi two-wheeler team
Project Introduction: With the increasing number of users and vehicles of shared two-wheelers, chaotic parking of vehicles not only affects the city appearance, but also endangers traffic safety. Therefore, orderly parking is the focus of the sustainable and healthy development of shared two-wheelers. Orderly parking usually relies on satellite positioning guidance, and its accuracy is usually 2-10 meters. However, most of the shared two-wheelers face urban canyons with many tall buildings, and the positioning accuracy and stability will further deteriorate, often causing users to fail due to positioning errors. The dilemma of returning the car seriously affects the user experience. In response to such problems, through in-depth cooperation with Wuhan University, we have gradually applied centimeter-level high-precision positioning algorithms. After three phases of cooperation, we have successively completed the construction of self-developed high-precision positioning algorithm from 0 to 1, the construction of tightly coupled positioning algorithm, and the evaluation of positioning performance under the condition of no real value, forming a complex system for two-wheelers. The positioning system of the scene, and continue to empower the two-wheeler business. In the practice of Qingju's credo of "creating user value with a long-term heart", we will gradually realize the mission of "making travel better".
Voice separation for in-vehicle environments
Project team: Professor Wang Longbiao of Tianjin University and Didi Intelligent Voice Interaction Team
Project introduction: Voice interaction is the core functional module in intelligence. Overcoming complex acoustic interference such as wind noise, tire noise, and multi-person dialogue, and distributing human voices in different positions without distortion is a necessary condition for smooth voice interaction. This cooperation with the team of Mr. Wang from Tianjin University has solved the problem that many people in the car are talking and disturbing the interaction, and realized the pre-research of the four-tone area separation algorithm. Our self-developed algorithm has low requirements on the stability of the hardware, and has good robustness to the acoustic environment, and its performance has reached a practical level under dynamic conditions. 2 academic papers have been submitted to the Speech Summit.
Research and development of data-driven decision-making control technology for high-level autonomous driving and its actual vehicle application
Project team: Professor Li Shengbo from Tsinghua University and Didi's autonomous driving team
Project introduction: Decision planning is the "brain" of autonomous driving vehicles. How to make vehicles smarter, safer, and more efficient, and improve the real-time calculation and scene scalability of decision planning are one of the main challenges faced by high-level autonomous driving. This project aims to explore the cutting-edge technology to realize the self-evolution of the "brain" of autonomous driving through data-driven training methods, so as to solve the dual problems of poor online real-time performance and poor scene adaptability of the original decision-making control scheme. The cooperation results include: proposing a complete set of data-driven decision-making control technology solutions; completing the real vehicle test of the first brain-inspired learning automatic driving system in China; developing 2 sets of core training and simulation tool software; publishing a number of top conference papers, including 1 won the best paper award of the top international conference; jointly trained a number of master and doctoral students, 2 of whom joined Didi Autopilot after graduation.
Award-winning teacher roundtable forum scene
Zhang Bo shared his thoughts on Didi Technology and industry-university-research cooperation in his opening speech. Zhang Bo said that in the past ten years, Dongfeng has built a shared travel platform with the help of the mobile Internet and smart phones. In the next decade, with the development of general artificial intelligence technology, Didi platform will bring huge opportunities for upgrading, travel safety and Efficiency will also be further improved. Faced with the huge demand of the academic community for industrial scenarios, big data, and computing infrastructure, Didi looks forward to promoting the transformation of scientific research results into products through cooperation and providing users with user value. Through continuous iterations over the past six years, this cooperation system has taken shape. In the future, Didi will further strengthen cooperation with the academic and research communities, and cultivate more talents in the cooperation, so as to promote the ecology of industry-university-research cooperation to become better and better, and to create more and more value.
Zhang Bo delivered a speech
According to Didi's business scenarios and technical characteristics, two sub-forums, "Down to Earth" and "Looking at the Starry Sky", were set up, and five award-winning project teams were invited to conduct project roadshows. Afterwards, the cooperating teachers and Didi colleagues talked about the possible changes in shared travel and online car-hailing ecology in the future, and how to use The power of industry-university-research cooperation was exchanged and discussed. In the future, Didi will continue to unite the forces of universities to continuously innovate and make breakthroughs, and explore the infinite possibilities of open innovation.
／ Didi Gaia Scientific Research Cooperation Project
Through the organic combination of cutting-edge academic research and actual business scenarios, together with the academic community to discover and define problems, cooperate and win-win to solve problems in the field, with the goal of promoting cutting-edge technological innovation and solving industry challenges, and realizing the transformation of scientific research innovation into industrial value. Since 2017, research topics in machine learning, knowledge mapping, smart transportation, geographic information technology, economics and other research directions have been opened successively, and together with researchers from various fields and Didi researchers, they have promoted the application of cutting-edge technologies in the field of travel. , Drive technological innovation with real scenarios.