How far are we from autonomous driving?

How far are we from truly large-scale implementation?

The pilot is just the beginning. How far are we from truly autonomous driving? Industry insiders believe that large-scale commercial applications still face many challenges, such as the urgent need for breakthroughs in some key technologies and incomplete supporting facilities.

Hangzhou Autonomous Driving Shuttle Demonstration Line "The most important thing about autonomous driving is to ensure safety. Even if 99% of road scene problems can be solved technically, it is still impossible to achieve true unmanned driving. The remaining 1% is the difficulty, which is the so-called 'long tail problem'." Wu Gansha, chairman and CEO of UISEE Technology, said in an interview with a reporter from China Electronics News that the difficulty of this problem lies in the fact that it is difficult for technology to exhaust all possible scenarios and cases, and it is difficult to define the full set of problems. If even the problem is not fully defined, it is impossible to effectively propose a solution, let alone effectively estimate the effect and cost of completing it.

Perhaps, as cars have made great progress in the coordination of intelligent algorithms and related hardware systems, the 1% problem can be gradually solved. "However, the import of high- computing-power computing chips and processing equipment in China is currently blocked. At the same time, manufacturers of vehicle-mounted computing platforms and operating systems in China mainly focus on secondary system development, lacking a tool chain for the entire process such as architecture design and system design. There is a risk that related core technologies will be controlled by others," said Qi Tianyu, vice president of the Autonomous Driving Western Science City Intelligent Connected Vehicle Innovation Center (Chongqing) Co., Ltd.

The commercial safety application of autonomous driving still requires more data accumulation and scenario training. Zhang Yaqin, an academician of the Chinese Academy of Engineering, said that Baidu's L4 autonomous driving test vehicles have traveled nearly 100 million kilometers in total, but data on extreme conditions is still insufficient, and there are challenges in data classification, labeling, processing, and compliance. At the same time, high software and hardware costs have also become an important obstacle to the promotion and function upgrade of autonomous driving. Yang Guiyong, an expert at the Technology Management Center of Geely Auto Group, believes that "the hardware costs of cameras , lidars , processors , etc. have declined in recent years, but they still significantly increase vehicle costs; the upgrade of autonomous driving functions will require higher hardware redundancy backup, increase the number of hardware and code in vehicles, and further increase costs."

In addition, in terms of infrastructure such as "vehicle-road collaboration", an effective commercial investment and operation model has not yet been formed. At present, local governments mainly rely on government funding, with high construction costs and weak sustainability, resulting in slow overall progress. The ultimate goal of autonomous driving is to make road traffic safer and make people's travel more convenient and efficient. The achievement rate of these two goals cannot be compromised. On-road pilot projects are a key step in the commercialization of autonomous driving, and it is more important to take this step steadily than quickly.