share_log

智己“漏看”、理想“错看” 边缘场景覆盖不足致高阶智驾事故频发

High-end smart driving accidents are frequent due to insufficient coverage of marginal scenes where wisdom “missees” and “missees” ideals

cls.cn ·  May 18 09:31

① “Zhiji is equipped with a parking system composed of 12 ultrasonic radars and 4 360-degree surround view cameras. When the square column is in the 45 degree direction, the ultrasonic radar will miss recognition, and the surround view camera's thin column recognition algorithm needs to be improved.” Zhu Xishan believes that the above accident is technically a marginal scene where it is difficult to recognize a thin square column. ② Zhu Xisan said bluntly that “autonomous driving” products that have already been mass-produced are still in the assisted driving stage.

Financial Services Association, May 18 (Reporter Zhang Yipeng) At a time when advanced intelligent assisted driving is rapidly becoming popular, safety accidents caused by this have occurred frequently.

On May 17, Wang Kang, project director of the Zhiji Auto Autonomous Driving Center, published an article stating that the APA parking trajectory was good after the optimization software was installed, and that the perception and detection of the “I-shaped metal column” was stable. When the system determined that it was close to the column, collision avoidance braking and path replanning were carried out in a timely manner. After more than 60 field tests, the Zhiji LS6 has completed obstacle avoidance and parking. “Based on the full-day testing situation yesterday (May 17), the optimization of this parking function has basically met the expected requirements. In the future, the IMOS 2.8.0 system will be installed and promoted to users along with other optimization content.”

Despite this, Wang Kang did not deny the problems with the Zhiji APA parking system itself. The day before, Wang Kang said bluntly that the thin column in the accident scene where the Zhiji user used the APA parking system to reverse the vehicle and enter the warehouse had a “work” shaped metal structure, and the phase size was too small. As a result, the position accuracy of obstacles constructed by sensors such as on-board ultrasonic radars and cameras was insufficient, causing the vehicle to collide with the thin metal column. Wang Kang also said that he will bear the user's maintenance costs at his own expense.

“Zhiji is equipped with a parking system composed of 12 ultrasonic radars and 4 360-degree surround view cameras. When the square column is in the 45 degree direction, the ultrasonic radar will miss recognition, and the surround view camera needs to improve the thin column recognition algorithm.” Zhu Xishan, a professor at the School of Automobile Engineering at Tongji University, believes that the above accident is technically an edge scene where it is difficult to recognize a thin square column.

Compared to the safety hazards of the Zhiji LS6 in automatic parking scenarios, the “ideal L9 high-speed rear-end accident” that occurred recently received more attention. The video shows that an ideal car, the L9, is driving on the highway with the assisted driving function turned on. During this period, the system detected a picture of the minivan on a billboard high in front, and suddenly braked, causing the rear car to rear-end. In response, Ideal Auto's customer service staff stated that the vehicle in question was the Ideal L9 Pro. This model uses a camera and millimeter-wave radar to identify obstacles in front of the vehicle and provide parameters for the driving assistance system. Limited by the limitations of current sensor recognition capabilities, the recognition may be inaccurate in some scenarios.

Although the two safety accidents mentioned above are different scenarios, they triggered the same “cold thinking” about advanced intelligent driving — how to cover more complete edge scenarios at a time when the installation rate of advanced intelligent driving systems continues to increase rapidly, and the determination of legal liability after the accident occurred.

Before L3 arrived, rhetoric such as “L2, L2+, L2.99999” abounded. Zhu Xishan told reporters bluntly that “autonomous driving” products that have already been mass-produced are still in the assisted driving stage, and there is no L2+ in the degree of automation classification for intelligent driving. NOA is a technology developed by enterprises according to L3 level autonomous driving. Although the hardware has the ability to achieve autonomous driving, the software also needs to continuously collect edge scene data for continuous iterative optimization.

On the issue of “the person responsible for the accident” that the industry is very concerned about, although Tesla CEO Musk promised that “if safety accidents occur due to FSD in the future, Tesla will bear full responsibility and pay maximum compensation to car owners,” judging from domestic laws, the driver is still responsible for traffic accidents caused by the NOA system. At the same time, Zhu Xisan emphasized that the Zhiji automatic parking accident was different. Although the Zhiji side handled it positively and stated that it would reimburse the parking accident repair costs at its own expense, from a legal perspective, automatic parking is a personal application, and the company is not responsible.

The translation is provided by third-party software.


The above content is for informational or educational purposes only and does not constitute any investment advice related to Futu. Although we strive to ensure the truthfulness, accuracy, and originality of all such content, we cannot guarantee it.
    Write a comment