Safety in intelligent driving: AEB becomes the focus

In the intelligent driving system created by Huawei, "providing consumers with sufficient safety functions" is the biggest selling point. In the luxury era when multi-sensor fusion 5R11V1L (5 millimeter-wave radars, 11 cameras and 1 lidar) is gradually becoming standard, AEB (automatic emergency braking) seems to have become a part of automotive safety technology.

Of course, we also see that the AEB level of some car models is still stuck in the era of one camera + one millimeter-wave radar (1R1V).

Let’s sort out some of the challenges of AEB.

01

Chinese car consumers often have a one-sided understanding of AEB (automatic emergency braking). With this wave of publicity, they may believe that AEB will automatically brake whenever and wherever an obstacle appears ahead.

However, if this is really done, the requirements for this system are too high. The triggering conditions of AEB and its performance in different scenarios are complex. (Automakers are reluctant to enumerate all the weaknesses in every case, so they are often ignored in one sentence, and consumers cannot deeply understand the vehicle manual.) This leads to a one-sided understanding of AEB (dim environments lead to reduced recognition capabilities, such as dawn, dusk, night, tunnels, etc.)

The passenger car AEB (automatic emergency braking) standard GB/T 39901 is a recommended national standard. C-NCAP evaluates the safety of new cars, including AEB functions.

● GB/T 39901 divides AEB into two sub-functions: collision warning and emergency braking, requiring warnings and braking to be issued in specific situations.

● C-NCAP evaluation includes positive response tests and false response tests for vehicles, pedestrians and two-wheelers.

The AEB test content has been gradually improved, but there are still some "weaknesses", such as limited test scenarios, insufficient test speed, and single weather conditions. Testing and improvements need to be carried out in more complex scenarios.

In fact, it is very difficult to make a perfect AEB. The unspeakable secrets of this function mainly include:

● Limitations on triggering target types: Currently, most AEBs only support responses to vehicles, pedestrians and two-wheelers, but not other target types such as human-powered tricycles, agricultural tricycles, animals, etc.

This is mainly because AEB requires accurate identification of target types, otherwise the false trigger rate will increase. Improving the accuracy of target identification requires a large amount of training data, and data on vehicles, pedestrians, and two-wheelers are the easiest to obtain, so the types of targets that AEB triggers have always been limited.

● Setting the braking speed: AEB’s braking strategy is usually determined based on factors such as vehicle speed and distance, but in high-speed driving scenarios, AEB often sets a deceleration threshold and releases the brakes when it is reached to avoid chain rear-end collisions.

This strategy may be fine in the era of assisted driving, but in the era of autonomous driving, how to continuously brake in high-speed driving scenarios without causing other collision risks is a technical challenge.

02

False trigger and missed trigger problems

There are two issues with AEB being publicized, mainly two false triggers:

● False triggering means that AEB is triggered when there is no risk of collision ahead, which not only affects driving comfort, but more importantly, driving safety. Some chain rear-end collisions are often caused by emergency braking of the vehicle in front. Reducing the false triggering rate has become one of the goals pursued by car companies.

● Missed activation: AEB is not activated when there is a risk of collision (missed activation will cause the driver to be more cautious), but considering that AEB is only an auxiliary rather than a dominant function for the driver, a certain degree of missed activation is acceptable. Car companies usually adopt the strategy of "better missed than wrong".

Therefore, from the perspective of car companies, reducing the false trigger rate is a direct goal, and the AEB trigger parameters are set very strictly, even to the point where it is difficult to trigger. A relatively mild approach, if there is enough time, is to avoid developing collision scenarios that are unlikely to trigger AEB, in order to reduce the possibility of false triggering.

AEB has very strict requirements for the screening of trigger targets, including identifying the target type first and ensuring that the recognition accuracy reaches a certain level. This is because different types of targets require different activation speeds, deceleration strategies, braking strategies, and prediction strategies, and if they are not distinguished, AEB will not be able to perform optimally. Moreover, in some cases, if the target type is not distinguished, directly triggering AEB may greatly increase the false trigger rate.

After being activated, AEB does not brake until the car stops. Manufacturers usually set a deceleration threshold, and once the threshold is reached, the brakes will be released. This is mainly because in high-speed scenarios, continuous braking may cause the vehicle behind to follow too closely, which may lead to a chain rear-end collision. In high-speed scenarios, once AEB releases the brakes, the driver needs to decide again whether to continue braking or choose to turn.

summary