This project is jointly developed by a university's intelligent transportation laboratory and multiple enterprises. It aims to address the problems of low decision-making accuracy and slow response speed in the field of intelligent driving under complex road conditions. After 5 years of technical research and development, it has been completed. The project has constructed an environment perception model that integrates multi-sensor data and uses deep learning algorithms to achieve real-time and accurate recognition of the surrounding environment of vehicles, with a recognition accuracy rate of 98.7%; We have developed a dynamic decision-making system based on reinforcement learning, which can complete decision-making and judgment in complex traffic scenarios such as intersection right of way allocation and sudden obstacle avoidance within 0.1 seconds.  

At the appraisal meeting, a appraisal committee composed of 7 industry experts listened to the project team's work report, technical report, and novelty search report, and observed the system's actual vehicle testing on site. Experts unanimously believe that the system has reached the international advanced level in multi-sensor fusion algorithms, dynamic decision logic, and other aspects. Some core technologies are in a leading position internationally, breaking the technological monopoly of foreign countries in the field of high-end intelligent driving decision systems.  

The project results have been applied to the development of intelligent driving prototype vehicles for three car companies, with a cumulative completion of over 100000 kilometers of real vehicle testing. The success rate of decision-making in complex scenarios such as rain, fog, and congestion has reached over 95%. 28 invention patents have been applied for related technologies, of which 15 have been authorized, 32 SCI papers have been published, and 3 enterprise standards have been formulated. It is expected that after production, the research and development cycle of intelligent driving systems can be shortened by 30%, and costs can be reduced by 25%.