By MIIT, CAST, CMIF as the guiding unit, China Association of Automobile Manufacturers hosted the "2023 China Auto Forum", held in Shanghai Jiading from July 5 to 7, 2023. Li Xiaoying, general manager of our company, was invited to attend the theme forum of  New Trends and Future Development of ICV Industry and participate in the roundtable dialogue.

With a history of 12 years, the China Auto Forum is the "Davos" of the global automotive field, the "weather vane" of the development of the automotive industry, and an important platform for exporting wisdom and ideas to the whole industry every year. Among them, the theme forum on the new trends and future development of ICV industry held on July 6 mainly focused on the direction of intelligent connected automobile market prospects, development paths, vehicle and road collaborative empowerment, intelligent travel services, and the integrated development of intelligent transportation and smart city, and jointly studied and promoted the technological development and innovative application of intelligent connected automobile industry. To help build an open and integrated new ecology of the automobile industry.

Li Xiaoying, general manager of Digauto, Chief Engineer of Dongfeng Technology Center, Chief Technology officer of Schaeffler China, Chief executive officer of Thunderxauto, and Vice president of Pony Ai were invited to participate in the roundtable dialogue on a series of issues to carry out in-depth dialogue exchanges and ideas.

On the round table dialogue, Professor Zhu Xichan asked General Manager Li of our company， "The automatic driving system of AI algorithm needs to carry out scene coverage test, compared with the typical scene test of ADAS, the test workload is greatly increased, accelerated testing is the key technology of the development of automatic driving system, what innovative methods are there of Digauto?"

Li Xiaoying, “The automatic driving system of AI algorithm first needs a large amount of scenarios data to cover the scenarios. For example, many enterprises now collect a large number of actual scenarios on the road, but after a period of collection, accumulated to hundreds of thousands or millions of kilometers, the probability of collecting effective scenarios will become lower and lower, and the efficiency of scenarios collection will become worse and worse.

We need to find a way to quickly improve scenarios coverage. The root cause of the rapid decrease of road collection benefit is the randomness of time and space in the appearance of actual traffic danger scenarios. Danger happens all the time, but the acquisition vehicle may not be exactly there at that point, and this time and space encounter is too difficult.

However, based on our understanding of the physical performance of the perceptual system, and the impact of environmental factors on the perceptual system, we can quickly find the types of dangerous scenarios. Then, these dangerous scenarios are reproduced through scenario construction, and these dangerous scenarios are directly collected by scenario acquisition vehicle. This way we can quickly increase the coverage of the scenarios.

In addition, it is calculated that the cost of obtaining raw data for a single dangerous scenario will be very low and more efficient.

In this way, human cognition and experience are well applied to the expansion of AI training data, which can accelerate the training process of AI.

In addition, for some dangerous driving behavior scenes such as forced cut-in lane change, emergency braking, etc., the original scenario data can also be obtained by means of construction.”

Digauto proposed to simulate the interference of natural environment factors and reconstruct the dangerous traffic flow scenarios under indoor conditions. We call it the vehicle traffic environment hardware in the loop test system (VTEHIL), which is mainly for the test and verification of intelligent driving functions. Now it seems that it can also be applied to the generation of raw data in AI autonomous driving scenarios.

In the early stage, we selected different types of vehicles and conducted corresponding tests under the indoor scenarios conditions we constructed. Through the tests, we found that the perception and decision-making of vehicles under the indoor scenarios conditions were highly consistent with that in the actual road environment.

In fact, some of the means and methods we now use for closed site testing are also the ideas of scenarios reconstruction. It's just setting up a scenario on the road.

At present, the VTEHIL lab has a prototype in Kunshan. The first lab is the conversion of the original Automobile Crush Laboratory into Automotive Autonomous Driving Safety Lab at Tsinghua University, and the construction of a full-scale indoor lab for complex natural environment simulation and physical traffic flow at the OEM in northeast China. Welcome all experts come there to guidance.

At present, the domestic intelligent connected vehicles are entering a new stage of rapid development, and the company will continue to deepen and actively promote the research and practice of intelligent driving active safety testing; Work together with the industry, move towards innovation, and help create a new ecology of development and integration of the automobile industry!