Although the investigations into these two incidents have not yet yielded final results, these incidents have still sparked heated discussions about the safety of autonomous driving due to Tesla's important position in the field of autonomous driving.
Autonomous driving still requires extensive testing and validation.
Crashing into cardboard boxes, dummies, traffic barriers… test videos of self-driving cars often leave viewers with lingering fears. While liberating humanity, the safety issues of self-driving cars also present corresponding challenges.
In recent years, my country's autonomous vehicle industry has developed rapidly. The China Intelligent Connected Vehicle Innovation Center predicted that from 2020 to 2025, the sales of intelligent connected vehicles that can achieve partial or conditional automation will account for more than 50% of the total vehicle sales in each year.
Safety is the cornerstone of the development of the autonomous vehicle industry. Experts and industry insiders say that while researchers are constantly improving autonomous driving technology, large-scale road testing and practical applications are needed to gradually improve the safety and stability of autonomous driving technology.
"From a technical perspective, the recent frequent accidents involving autonomous vehicles can all be attributed to vulnerabilities in the autonomous driving system," said Professor Xie Hui, director of the Interdisciplinary Research Center for Autonomous Vehicles at Tianjin University. He believes that these problems are inevitable during extensive application testing of autonomous vehicles. "No matter how perfect the algorithms for perception, decision-making, planning, and control in autonomous driving applications are in the laboratory stage, vulnerabilities may still emerge."
Shi Jianping, Vice President of Intelligent Driving at SenseTime's Mobile Intelligence Business Group, stated that although researchers have been continuously iterating on the reliability, safety, and redundancy of intelligent driving software and hardware systems, there are still many unknown issues to be addressed for autonomous driving systems.
Since the world's first modern automobile appeared in 1886, the automobile has continuously exposed various problems in its application and development over the past century. At the same time, through continuous improvement, its technology and safety performance have gradually become more stable. Xie Hui said, "Compared with traditional automobiles, autonomous vehicles are far behind in terms of quantity, usage time, and experience in dealing with different scenarios."
Regarding the issue of autonomous vehicles seemingly ignoring obstacles, Shi Jianping believes this may be due to a problem with the perception module. "The perception module is like the eyes of an autonomous vehicle; if the eyes malfunction, safety is out of the question," Shi Jianping stated. "Therefore, in the past two years, perception technology has been the fastest-growing and most significantly improved of the four core modules of autonomous driving."
In Shi Jianping's view, autonomous driving is a highly complex system involving the interaction and coordination of numerous sensors. The reliability of each piece of hardware affects the overall reliability of the product. Furthermore, adjustments to the types and quantities of equipment can lead to an exponential increase in the risk of product unreliability, requiring extensive testing and verification.
Improve laws and regulations to safeguard road testing
Road testing is an essential stage in the maturation and improvement of autonomous driving technology. Dr. Yang Ning from the School of Law at Tianjin University explained that current autonomous driving testing mainly consists of two types: road testing and simulation testing. Many companies are developing their own simulation testing platforms to improve the efficiency of technology iteration, verify product reliability, and anticipate potential risks. However, with current technology, simulation environments can hardly realistically reproduce actual road conditions; therefore, simulation testing can hardly replace actual road testing.
“Conducting road tests inevitably involves questions such as whether autonomous vehicles need to apply for licenses to be on the road and who will be responsible in the event of a traffic accident…” Yang Ning pointed out that if these issues are not resolved, they will hinder the development of autonomous driving technology and impede the commercialization of autonomous driving.
Internationally, countries such as the United States, Germany, and South Korea are accelerating the development of legislation related to autonomous driving. For example, in 2017, the United States approved the Autonomous Driving Act to encourage the testing and development of autonomous vehicles; in 2018, Germany revised its existing road traffic law and issued the world's first code of ethics for autonomous driving; and in 2020, the South Korean Ministry of Land, Infrastructure and Transport released guidelines for the safe operation of autonomous vehicles.
In April of this year, my country's Ministry of Public Security drafted the "Road Traffic Safety Law (Revised Draft)" (hereinafter referred to as the "Revised Draft"), which clarifies the relevant requirements for road testing and traffic of vehicles with autonomous driving functions, as well as the provisions on the apportionment of liability for violations and accidents. This is the first time such a draft has been produced at the national legislative level and is expected to fill the legal gap in the autonomous driving industry.
Yang Ning stated that my country's development of autonomous driving has long lacked the guidance of relevant laws and regulations. The revised draft includes provisions for autonomous vehicles, which will undoubtedly greatly promote the development of autonomous driving.
"The revised draft clarifies the legality of autonomous vehicles entering public roads, providing a solid legal basis for the practice of road testing of autonomous vehicles in various regions," said Yang Ning.
According to incomplete statistics, more than 20 cities in my country have issued policies related to autonomous driving road testing. On March 23 this year, Shenzhen took a leading step in local legislation, releasing the "Shenzhen Special Economic Zone Intelligent Connected Vehicle Management Regulations (Draft for Public Comment)". Its content covers the entire chain of intelligent connected vehicles, including road testing and demonstration applications, access and registration, use management, network security and data protection, vehicle-road cooperative infrastructure, road transportation, traffic accident and violation handling, and legal liability. On April 10, the overall implementation plan for the Beijing Intelligent Connected Vehicle Policy Pilot Zone was approved, which will open up highway testing for autonomous vehicles to accelerate the implementation of autonomous driving applications such as highway trunk logistics.
However, Yang Ning also stated that the provisions on autonomous driving in this draft revision are only general principles, and he hopes that more specific provisions will be added during the subsequent review.
Improving security requires dual oversight of technology and standards.
Xie Hui believes that in addition to extensive testing and verification, breakthroughs are urgently needed in technologies to improve the safety of autonomous vehicles. "The most pressing breakthrough is in patching security vulnerabilities. This requires first designing testing methods to expose algorithmic vulnerabilities as much as possible. In addition, a relatively independent safety operator module needs to be designed within the autonomous driving control system. This 'safety operator' monitors the operational safety of the entire system, and can issue timely warnings and remind the system to take emergency measures once a vulnerability is discovered."
"For example, autonomous vehicles rely heavily on GPS, but GPS signals need to be transmitted to the vehicle via satellite, and signal distortion or loss may occur during this process. If there is a safety module that monitors the quality of GPS signal indicators, once an abnormal GPS signal is detected, another signal can take over to ensure that the vehicle does not go out of control," Xie Hui explained.
"In addition to technological breakthroughs, we should also establish a complete set of management systems for the safe testing and application of autonomous vehicles, a new technology," Xie Hui said. "Autonomous vehicles are like children. After teaching them various skills, we also need to set rules for them to ensure that they do not make mistakes."
"Taking an autonomous driving engineering vehicle as an example, a complete set of management systems must be established before its application. This includes process design, construction supervision responsibilities, dispatching personnel's job responsibilities, accident handling mechanisms, and maintenance mechanisms. A complete set of systems is needed to support the operation of the autonomous driving system." Xie Hui believes that "currently there is no mature management system that can be applied, and we still need to explore and improve it continuously."
Shi Jianping also believes that "every module of an autonomous vehicle may have safety hazards. The core of solving these problems is to formulate standards and industry norms, and to evaluate the safety, reliability, and stability of each company's autonomous driving system according to these standards and norms." She explained that the current industry-standard autonomous driving license exam assesses whether an autonomous vehicle can perform certain operations under autonomous driving conditions by setting different scenarios, mostly focusing on functional evaluation. However, in reality, there is no complete standard for this evaluation. For example, how many scenarios must the system pass stably and reliably over a long period of time to mean that the driver can be gradually removed from driving responsibility and no longer participate in the autonomous driving system? There are still gaps in standards and regulations in this regard.
