Sunday, 2022 December 4

Who takes the blame in an autonomous vehicle accident?

It’s full speed ahead for the autonomous vehicle industry, but as more of these vehicles take to the roads, it’s becoming clear that regulatory frameworks need a lot more expounding.

Recent incidents have sparked safety concerns, including an accident in mid-August where an XPeng P7 collided with a stationary broken-down vehicle on an overpass in Ningbo, Zhejiang. The P7 had been moving at a speed of 80km/h with the LCC (lane-centering control) assisted driving function activated, and the system failed to detect the broken-down vehicle in front. The driver also had no time to retrieve control of the P7. Unfortunately, the accident resulted in the death of the man who had been standing in front of his broken-down vehicle at the time. The driver of the P7 suffered minor abrasions.

A spokesperson from XPeng said they would fully cooperate with relevant departments to investigate the accident.

So, which party is responsible for such an accident? Is it XPeng, the company that manufactured the car and its autonomous system? Or is it the driver of the P7?

A lack of clear, standardized legislation

According to the Road Traffic Safety Law of the People’s Republic of China, the rear vehicle is fully responsible for rear-end collisions.

Should a vehicle break down on an expressway, a warning triangle must be set up 150 meters behind the vehicle. In weather conditions that result in poor visibility, such as rain and fog, the distance must be increased to 200 meters.

In this particular accident, the owner of the stationary vehicle had not placed the requisite warning triangle behind his car. Whether it would have made a difference, however, is unclear.

The fact that the P7 was in assisted driving mode makes things complicated since legislation for autonomous vehicles has yet to be standardized in China.

The most comprehensive legislation on autonomous vehicles thus far had only been formally implemented weeks ago on August 1, 2022, titled Regulations on the Administration of Intelligent Network-Connected Vehicles in Shenzhen Special Economic Zone.

The regulations include provisions on access registration and road use of intelligent network-connected vehicles. They also provide the first crucial framework for assigning liability in autonomous driving accidents. In summary, liability is assigned to the driver if he was driving the autonomous vehicle. If no one was driving, then the liability falls to the owner of the vehicle. Should the accident be due to vehicle defects, the owner can claim compensation from the vehicle manufacturer.

This legislation in Shenzhen is the most explicit regulatory framework for autonomous vehicles in China so far. While several other cities in China have issued guidelines or implemented rules on self-driving test vehicles on the road, they do not specify the assignment for liability for traffic accidents in detail.

In other words, it’s unclear whether the driver of the P7 or XPeng should be the one to take responsibility. However, if the ruling follows the current Road Traffic Safety Law in China, then the responsibility falls squarely on the driver’s shoulders.

The level-based system in other countries

In other countries, a level-based system is used to determine liability when it comes to autonomous vehicles:

  • (L0) No Driving Automation — The driver controls the vehicle, including braking, steering, throttle, and transmission. The driver must assess the danger of the situation at all times.
  • (L1) Driver Assistance — The car can support the driver through features such as a lane-keeping system, an automatic braking system, and an adaptive cruise function.
  • (L2) Partial Automation — The system can control the steering as well as acceleration and deceleration. The driver sits in the driver’s seat and can take control of the car at any time.
  • (L3) Conditional Automation — The autonomous driving system is equipped with environmental detection capabilities and can operate the vehicle, but human override is still required.
  • (L4) High Automation — The system fully controls the vehicle and assesses the traffic environment in the process. No passengers are required to control the vehicle at any time, but there is still a manual override option. Can only be used in cities or on highways.
  • (L5) Full Automation — The vehicle performs all tasks, and the user does not need to control the vehicle at all in any scenario. In other words, this refers to “full automatic driving” or “unmanned driving.”

The XPeng P7 belongs to the L2 category of partial automation. Its system is responsible for steering, accelerating, and decelerating, while the driver is responsible for other driving operations. Since the driver is responsible for monitoring the environment and responding to failures, a qualified driver must sit in the autonomous driving vehicle and keep their hands on the steering wheel.

The liability then depends on the cause of the accident. If the cause were an operational error in steering, acceleration, or deceleration, both the system and driver would be responsible, but the division of responsibility may vary. Ultimately, the driver is responsible for monitoring the driving conditions and correcting system errors. However, if the system does not allow for the correction of errors, then the system operator is liable.

In the P7 accident, the driver failed to apply the brakes in time. Although this was due to both the system’s failure to issue a warning and the driver’s own distraction at the time, the primary responsibility falls on the driver.

Current autonomous driving systems

Therein lies the question: what are autonomous vehicle manufacturers doing to minimize potentially fatal system errors? And what are the current measures in place?

Autonomous vehicles use technology such as millimeter-wave radar and LiDAR (light detection and ranging) sensors to detect objects around the vehicle. According to XPeng’s official website, the P7 is only equipped with millimeter-wave radar.

The biggest disadvantage of millimeter-wave radar is that it is not ideal for stationary object recognition. As the name suggests, this technology is based on millimeter waves (electromagnetic waves with wavelengths between 10 mm/30 GHz and 1 mm/300 GHz) and can detect the location of stationary objects.

However, it cannot accurately identify the types of objects—leading to “phantom braking” for things like traffic lights and billboards that the car mistakes for obstacles. To resolve this issue, manufacturers employ static clutter filtering, which filters out the signals from stationary objects. This, of course, creates its own problems when there are real stationary obstacles on the road.

Combining millimeter-wave radar with a camera-based system helps, but a drawback of the latter is its poor performance in recognizing white objects, which is a major technical reason for many accidents.

However, the LiDAR sensors in the newer XPeng models released this year are an improvement on the current sensor architecture.

LiDAR was introduced to avoid collisions and accidents caused by situations that existing technologies cannot deal with, such as stationary vehicles with an unconventional shape. With LiDAR, stationary object recognition is no longer an issue.

An industry-wide issue

It seems autonomous vehicle manufacturers all over the world are facing similar challenges when it comes to clearly defining what their products can and cannot do.

Tesla recently came under fire from The California Department of Motor Vehicles (DMV) for disseminating misleading statements. The DMV said that Tesla had exaggerated its advanced driver assistance system (ADAS) advertisements with statements such as “no action required by the person in the driver’s seat” on the Autopilot page of its website. This, the DMV argued, misleads consumers into thinking that no driver supervision is required.

In August last year, after an accident involving a NIO vehicle with an assisted driving feature, all vehicle manufacturers switched to “assisted driving” instead of more misleading ones like “autonomous assisted driving,” “autonomous driving capability,” and “autopilot”.

Li Xiang, the founder of LI Auto, called on the media and automotive industry to unify the standard Chinese terms for autonomous driving with a system similar to the level-based one in use by other countries. To avoid misunderstandings caused by exaggeration in advertisements, simple and standardized labeling should be used. He also pointed out that “restraint in the promotion of autonomous driving and increasing investment in technology will benefit users, industries, and enterprises for a long time.”

Emphasizing the driver’s responsibility

The driver’s role—and responsibility—in autonomous driving is now in greater focus. NIO, LI, and XPeng users must now have proof of a driving license before they can activate the high-level assist function. There are also safety training tests that users must undergo, with mandatory acknowledgment of the manufacturer’s disclaimer that the ultimate responsibility lies with the driver.

Other monitoring measures have been put in place, like XPeng’s Smart Driving Points scheme, which was launched at the end of 2021. Points are deducted or added according to drivers’ habits and rule violations. If the driver takes both hands off the steering wheel, fails to take over the vehicle in time when required to do so, drives for an overly long time, or drives while fatigued, points will be deducted.

Such measures aim to help drivers use the autonomous assisted driving function with purpose and deliberation instead of viewing it as an autopilot feature that enables them to switch off mentally while the car takes care of the journey.

While there’s talk of a future with people zipping around in fully autonomous cars, a scenario like that is still far from being realized. For the moment, let’s call a spade a spade—these assisted driving functions are no more than a supporting feature. Human beings are still very much in the driver’s seat.

This article was adapted based on a feature originally written by Wang Shuo Qi and published on AutoR Smart Driving (WeChat ID: zhinengqiche). 36Kr Global is authorized to translate, adapt, and publish its contents.

KrASIA Connection
KrASIA Connection
KrASIA Connection features translated and adapted content published by 36Kr.com, the largest and most influential tech portal in Chinese language.
MORE FROM AUTHOR

Related Read