Author / Yu Lei
Recently, American Tesla driver Kevin George Aziz Riad was charged with manslaughter for abusing a driver-assisted driving system to cause death. It has also become the first case in the history of the United States and even the world to be charged with a crime after a death due to a driving assistance function.
The accident occurred in 2019, Tesla driver Riad did not effectively supervise the vehicle when the Autopilot function was turned on, because the Autopilot at that time did not support the identification of traffic lights, causing the vehicle to run a red light and crash into a normal Honda Civic, killing two occupants in the other car. At the time of the accident, the driver was not able to take over the steering wheel in the first place, which caused a tragedy.
The misuse of driver assistance has led to a number of tragic cases
In fact, in recent years, there have been many accidents caused by the abuse of driver assistance systems. According to relevant statistics, since 2016, the National Highway Traffic Safety Administration NHTSA has investigated 26 Tesla Autopilot-related accidents, which have killed at least 11 people, but most of the accidents have not failed the system, mainly due to consumers' over-trusted driver assistance systems.
Not only Tesla, but also other new car-making forces have similar accidents, such as last August, Lin Wenqin, founder of Meiyihao Brand Management Company, rear-ended the highway maintenance car in the case of using the ES8 NOP function, and unfortunately lost his life.
After such accidents, most of the voices in the society focus on the wrong driving assistance system of the car company, inducing the promotion of the automatic driving system, resulting in the abuse of the owner. But in reality, the lack of a reliable way for driver assistance systems to determine whether a driver is effectively supervising a vehicle is also a key problem.
"There is currently no evidence that driver assistance systems can make vehicles safer, but studies have shown that if the systems do not ensure driver involvement, their safety may be reduced." David Harkey, president of the IHS of the American Highway Safety Insurance Association, said.
Zhu Xichan, director of the Institute of Automotive Safety Technology at Tongji University, also pointed out: "For L2+ level assisted driving, the biggest problem it faces is that consumers expect it much higher than its capabilities. Although the ability of L0, L1, L2 level assisted driving systems is very poor, but the user's expectations for it are even worse, so it will not exceed their ability range in use, which makes these systems safe to use."
Under the current requirements of laws and regulations, the L3 level automatic driving system has not yet been able to be put into the civilian car market, and the L2+ driver assistance system continues to expand the high-end automatic driving scenario, which will be the development direction in the short and medium term. However, this approach leads to system supervision needs, which cannot grow together with the increase in the applicable scenarios of driver assistance systems, and will still stay in the L2 stage that requires full supervision, but drivers are likely to be over-trusted by the increase in scenarios.
U.S. Consumer Reports also says that when a task becomes automated, people stop putting effort into it. Just like after switching from a manual transmission to an automatic transmission, the driver's energy is put into places where there is no automation yet.
In addition to the abuse of driver assistance systems, accidents caused by inability to effectively supervise vehicles, such as inattention and drunk driving, for similar reasons, are even more, and this is one of the scenarios in which driver monitoring systems can be used. A set of data previously released by NHTSA showed that distracted driving caused more than 3,000 deaths in 2019, a 10% increase from 2018. Accidents caused by drunk driving are also on the rise, with 10,142 people dying in crashes while drunk in 2019.
Rely on vision solutions to help drivers focus on driving
At present, the driver assistance system requires the driver to maintain supervision in real time, but the way it confirms that the driver is supervising the vehicle is generally indirectly judged by sensors to determine whether the driver is holding the steering wheel, but this is often not directly related to the driver's state. Previously, when the driver assistance system was activated, the owner of the car deceived the driver of the supervision system into holding the steering wheel by hanging the water bottle on the steering wheel.
Compared to steering wheel sensors, the vision-based driver monitoring system (DMS) is considered a more promising monitoring technology. Such systems can more accurately determine the driver's state through visual cues such as facial, hand, or physical features recognized by the camera.
While this is a visual solution where you can see images inside the car, the driver monitoring system uses a closed-loop system, which means that all data remains on the side of the car. The system is also not to "denounce" the driver, the purpose is to help the driver drive more safely, the camera screen used to perceive the driver's state should not be saved and uploaded, nor should it cause any privacy problems for consumers.
In addition, there is an equally serious problem with driver monitoring systems, which still face the challenges of the perception layer, the cognitive layer and the judgment layer, how to accurately define fatigue, measure fatigue, and how to accurately measure the driver's attention to the road. As a system on standby for a long time, it is necessary to achieve a very low false positive rate. Moreover, most driver monitoring systems in the current phase can only recognize a single visual cue, which also leads to its poor robustness, such as when there is occlusion or light changes, it is easy to be interfered with.
In addition, relevant practitioners have proposed that with the widespread application of driver monitoring systems in the future, pure end-to-end deep learning neural networks are no longer enough. Safety-focused regulators are looking for reliable solutions to monitor drivers' status, ensuring they are engaged and able to regain control of vehicles if necessary.
Still, the potential of driver monitoring systems is there for all to see. U.S. Consumer Reports has said safety experts from groups such as the National Transportation Safety Board, the U.S. Road Safety Insurance Association IIHS, the European New Car Assessment Program and the Road Safety Insurance Association say that in order for drivers to remain in the decision-making circle, driver assistance systems should be combined with driver monitoring systems.
Among them, the U.S. Consumer Report and iiHS have made it clear that starting this year, the rating of the driver monitoring system will be a factor in the score of new models.
U.S. Consumer Reports will introduce evaluation criteria based on driver monitoring systems for the first time in the 2022 car preferred models announced on February 17, and the lack of qualified such systems will result in a loss of 2 points in the total score of the model, which will be expanded to 4 points by 2026. William Wallace, U.S. Consumer Reports Security Policy Manager, also stressed that they will retain the right to withhold points in accordance with automakers' privacy practices: "Strong privacy protections are non-negotiable, and car companies need to convince drivers that these cameras will only be used for their safety."
IIHS said on Thursday that a rating plan for their driver monitoring system is already in the works and is expected to be rolled out later this year, but that it could also be delayed if vehicle testing is not possible due to a global semiconductor shortage.
At this stage, most systems have limited usefulness
As of now, U.S. Consumer Reports has published the test details of some models. The results show that most driver monitoring systems are worthless, with only Ford and GM's systems rated as safe and fully functional. Most active driver assistance systems, including Tesla Autopilot, do not use effective direct driver monitoring, but instead determine whether the driver's attention is focused by indirectly identifying steering wheel pressure.
Among them, the details of the testing of Tesla Model Y and Model S by the US Consumer Report have been released at the end of last year. However, because the two later pushed v11.0 (2021.44.25.2) and FSD beta 10.8 (2021.44.25.6) updates, and retested, but the driver monitoring system is no different.
The U.S. Consumer Report noted that although tesla Model Y and Model S have in-car cameras, they also mentioned in last May's update that "the cabin camera above the rearview mirror can now detect and alert drivers to inattention when autopilot is enabled", but the actual regulatory role is very limited.
The test results show that the Tesla Model Y and Model S's in-car cameras are only used as a supplement to the steering wheel pressure sensor, reducing the time allowed to disengage only if the driver's line of sight is not on the road ahead. But if the steering wheel can feel the pressure, whether it's looking at the road ahead, moving your gaze to another place or playing with your phone, it's no different.
They also found that the two cars' in-car cameras were still allowed to use the Autopilot and FSD Full Self-Driving beta systems when they were obscured.
Kelly Funkhouser, technical manager of vehicles at Consumer Reports in the United States, said: "A qualified driver monitoring system should detect when the driver is inattentive and alert the driver to attention. If the driver does not respond to these alarms, the system should escalate the warning. If there is still no response, ideally, the system should make the vehicle stop as safely as possible."
During the test, U.S. Consumer Reports was also supported by an evaluation by researchers at the MIT Advanced Vehicle Technology Alliance (AVT), which also conducted similar tests against the Tesla Model 3. "On the Tesla Model 3 we studied, it's not clear how camera-based driver monitoring systems can support drivers." Bryan Reimer, a research scientist at MIT's AgeLab and deputy director of the MIT's Transportation Center at New England, said.
In addition, bmw and Subaru's driver monitoring systems have also given details of the missing points. BMW's Traffic Jam Assist driver monitoring system only works effectively at speeds below 40mph (about 64.37km/h) and can turn off the in-car camera individually or enable the driver assistance system when the camera is covered. Subaru EyeSight and DriverFocus can also be completely turned off via the menu, and the driver assistance system will work when it is turned off.
Of the test models, only the Ford BlueCruise and the GM Super Cruise scored. U.S. Consumer Reports said it's not just that they use a straightforward driver monitoring system, but also because of their infrared camera mounted on top of the steering column that can accurately constrain the driver to look at the road.
The Ford BlueCruise warns the driver with audio and visual alerts when the driver closes his eyes or looks off the road for about 5 seconds. If the driver does not respond, the vehicle briefly slams on the brakes to "alert" the driver into an alert state and eventually slows down to a stop; the universal Super Cruise can remind the driver to concentrate with multiple warnings. If the driver still does not respond, the system controls the vehicle to slow down and eventually stop it. If the camera is obscured, the system cannot function either.
Regarding the privacy and security of the cameras in the car, BMW, Ford and General Motors told U.S. consumers that data or video from their systems is only stored in the car. Subaru told U.S. Consumer Reports that their DriverFocus system did not record any information. The camera inside the Tesla car can record video, and if the driver allows it, the video can be sent directly to Tesla for its use.
Although from a technical point of view, the driver monitoring system is a transitional stage of technology, mainly in the L2-L3 level to play an important role, L4 level away from the safety officer, there is no longer a driving safety level of significance. However, under the current trend, the L4 level automatic driving system is still hopeless to enter the passenger car market in the short and medium term, and the penetration rate of L2+ driver assistance systems that are constantly expanding the scene will increase rapidly. At this stage, a reliable and efficient driver monitoring system is essential.
In China, the relevant departments are also strengthening consumers' awareness of the capabilities and needs of autonomous driving systems at all levels, such as the GB/T 40429-2021 "Automobile Driving Automation Classification" recommended national standard proposed by the Ministry of Industry and Information Technology in August last year, clarifying the functional operations and responsible subjects at all levels. At the same time, the recommended national standard "Driver Attention Monitoring System Performance Requirements and Test Methods" is also being formulated to guide the rapid and benign development of driver monitoring systems.
For reference:
https://www.consumerreports.org/car-safety/tesla-driver-monitoring-fails-to-keep-driver-focus-on-road-a3964813328/
https://www.consumerreports.org/car-safety/driver-monitoring-systems-ford-gm-earn-points-in-cr-tests-a6530426322/
https://www.consumerreports.org/driver-assistance-systems/ford-bluecruise-active-driving-assistance-system-review-a5728630415/
https://techxplore.com/news/2022-01-vehicles-drivers.html
About Auto Byte
Auto Byte launches the Automotive Technology Vertical Media for The Heart of Machines, focusing on cutting-edge research and technology applications in the fields of autonomous driving, new energy, chips, software, automobile manufacturing and intelligent transportation, and helping professional practitioners and relevant users in the automotive field understand technology development and industry trends through technology to gain insight into products, companies and industries.
Welcome to follow the stars and click Likes and Likes and Are Watching in the bottom right corner.