In today's increasingly popular L2 level driving assistance, many users have a certain dependence on the L2 level assistance system, in the high speed or fast driving or in the congestion time, the system can greatly reduce the driver's fatigue, so that driving more relaxed and comfortable. Since the L2 level of driving assistance is so good, then the L3 level system is more exciting, but why is the L3 level assistance system still not available?
In fact, the Audi A8 has implemented the L3 level system in 2017 and announced that it will be mass-produced in 2021, but the recent news is that they have announced that they have given up, and more manufacturers have said that they will skip the L3 and go directly to the L4 level. This fully shows that there are some difficult problems in the L3 level, the most obvious of which is that the L3 level itself is not clearly defined, and the division of responsibilities is even more undefinable.
In the definition of SAE in the United States, L3 level driving assistance is called conditional automatic driving, that is, the vehicle can achieve automatic driving in good road conditions, and when it is impossible to determine, it reminds the driver to take over. But this definition is obviously more vague, in the end which scenarios are under the automatic control of vehicles, when to hand over to people to take over, it is difficult to quantify specifically. The scene division is not clear, the responsibility is not clear, which is very risky for automobile manufacturers, and it is difficult for policy makers to start, so L3 has become an embarrassing existence.
However, although the definition in the L3 regulation is not clear, it does not prevent manufacturers from pushing the technology forward by defining it themselves. GAC BJEV has tried to define L3 and realize it through a series of new technologies. Their ADiGO system has been upgraded to version 3.0, with new upgrades on both hardware and chips. We recently test-drove the Aion LX 2020 to see how they understand and implement the L3.
1, vehicle perception resolution is not enough high-precision map to make up
We all know that for vehicles to achieve automatic driving, it is the basis for the acquisition of external information. In terms of perception, the combination of Tesla's 8 smart cameras is undoubtedly quite comprehensive, and the amount of information obtained is large enough to put forward higher requirements for GPUs (image processors) and chips, which is obviously not something that general car manufacturers can do. Most manufacturers use Bosch-based systems to use a camera + millimeter wave radar + ultrasonic radar solution, the advantage of this solution is that the amount of data for image perception is not large, but the weakness is that the resolution of the information is relatively low.
Aion LX's ADiGO 3.0 hardware is basically the same, it has 12 ultrasonic radars, 4 lateral millimeter wave radar, 1 forward millimeter wave radar, 4 panoramic cameras, and added 1 Mobileye front-looking intelligent camera composition, although the total number of sensors reached 22, for high-speed and fast straight-line driving, the road information is enough, but to achieve automatic vehicle line overtaking, it is still not enough.
Therefore, in the flagship intelligent driving suit (optional price of 39800 yuan) at the top of ADiGO 3.0, a positioning system based on high-precision maps has been added to realize the highway driving assistance HF, that is, when driving on high-speed and expressway roads, the vehicle can achieve: 1, completely open hands and feet to drive, the system will not remind the driver to hold the steering wheel in a short period of time, unless it encounters special circumstances; 2, when encountering a slow car in front, the vehicle will remind the driver according to the road conditions behind both sides. The driver dials the turn signal pole to issue a command, and the system automatically completes the overtaking of the line change.
ADiGO 3.0 uses high-precision maps from Baidu, with the basic map of the country and the high-precision map of the province where the vehicle is located, which is updated and updated every quarter. Compared with ordinary maps, high-precision maps have improved their positioning accuracy from 5 meters to 0.1 meters, which can ensure that the located lanes are completely accurate, while ordinary maps are difficult to identify which lanes the vehicle is driving on, which is the prerequisite for achieving intelligent lane changes. In addition, THE HIGH-PRECISION MAP has data support for the slope, bending curvature and roll angle of the road, so that it can help control the driving. Through intelligent cameras to identify traffic signs, for instant speed limits, road closures and various reminders can be recognized and reminded to the driver. It can also predict the road conditions ahead in advance and control the speed in advance to ensure the comfort and economy of driving through data comparison.
2. Mobileye Eye Q4 intelligent visual perception system
On the current level of visual perception suppliers, Mobileye is obviously in a leading position, the front-facing intelligent camera detection distance on this car is up to 200 meters, which can do any angle recognition, obstacle recognition and multi-target tracking in front of the car, and the Mobileye Eye Q4 chip also has strong computing power. As can be seen from the dashboard, the big cars and small cars in the left and right lanes of the front view can be intuitively displayed, in addition to the longitudinal view, it can also identify objects moving laterally, but whether it is pedestrians or cyclists, the display is a motorcycle, and the lateral movement of the picture is also very strange, and there is a lot of deviation from reality.
3. Driver monitoring
In high-speed and fast-moving autonomous driving, it is easy for drivers to be distracted by "nothing to do", but for the system, the driver is also responsible for monitoring and is ready to take over the vehicle. For vehicles, it must add a driver fatigue monitoring system to monitor the driver's status. This is also one of the embarrassments of L3, on the one hand, the driver wants to lazily rely on the vehicle to control, on the other hand, the vehicle is inseparable from the driver's monitoring, and in turn monitors the driver. The driver's facial state is monitored by the camera placed on the steering wheel, and the driver's state can be determined whether the driver is tired, on the phone, distracted (such as looking at the mobile phone), smoking, etc. through AI intelligent analysis, and then reminded. This set of monitoring system face and eye recognition is very sensitive, we in the use of the process of a little inattentive will sound an alarm, but some of the impact on driving, in the setting menu can be set to retain fatigue monitoring, the tolerance of the system will be much higher.
4. Steering wheel induction
Because autonomous driving supports both hands leaving the steering wheel, vehicles need to know if the driver's hands are on the steering wheel. The steering wheel on Tesla and other models uses a pressure sensor, which is integrated in the center to determine whether the hand is holding, such a weakness is that it is easy to cheat to deceive the system, such as many people put drinks or apples with transparent glue on the steering wheel, the system is difficult to identify. The steering wheel of the Aion LX uses capacitive sensors, which eliminate the possibility of the driver cheating the system.
Unfortunately, this test drive is not equipped with a flagship intelligent driving suit, there is no high-precision map support, and L3 level driving assistance cannot be realized, which is still L2 level. But we are still looking forward to ADiGO 3.0 and look forward to going into the test drive in the future.