After the ideal OTA, there is one data that equals Huawei.
In the latest OTA 5.2 version of Li Auto, the main improvement is the high-speed NOA capability. Along with the updated information, there is also its takeover rate, 1320 km/time. Last time, Yu Chengdong personally drove home for the New Year, and also said that he did not take over the whole process, which is almost 1314 kilometers. Looking at the data, the Ideal AD Pro/Max 3.0 and Wenjie ADS 2.0 are tied at high NOA.
Of course, there is no specific indication of the conditions under which the high-speed completion of the 1320 km/time takeover, the road section, traffic flow, road conditions, etc., do not give specific reference information. Then, the new function is the ability to bypass the cone barrel and the anti-collision barrel; The maximum detour speed for cone barrels and crash barrels is 80 km/h, and the maximum detour speed for stationary vehicles is 100 km/h. Let's probably know what features have been added and what the upper limit of the features is.
So, how does the post-OTA ideal come about? In terms of use, can it surpass Huawei?
If the hardware is not replaced, how can the perception be optimized?
For the optimization of the high-speed NOA function, in fact, the main central idea is to make the vehicle run as smoothly as possible under high-speed NOA conditions, and to make as few errors or unexecutable instructions as possible, so as to complete the use experience of no or less takeover under ultra-long mileage, such as about 1300km to take over once or without takeover.
Then the optimization points are mainly focused on the following aspects:
- Enhance the ability to make and execute lane changes, more accurately judge when and what path to use to achieve lane change, and improve driving safety and consistency;
- The improvement of safety avoidance ability is actually the ability to identify and avoid obstacles in the high-speed state and above 80km/h speed domain;
- For the detour and braking of stationary vehicles and vehicles with large speed differences, the system can now improve the detour and braking capabilities of stationary vehicles and vehicles with large speed differences in the same direction, and support the speed range of vehicles increased to 100km/h;
- The key point is that optimizing the timing and game strategy of lane change is a game on the strategy of changing lanes of the system, changing lanes in space and waiting, which can reduce the initiation time of lane change and overtaking, and improve the decisiveness and effectiveness of lane change decisions (more efficient);
- Then there are strategies such as defensive lane change strategies, bus lane identification and avoidance strategies, which also need to be optimized.
Then, the lowest level of the ideal optimization this time is the ideal AD PRO 3.0 system, which is realized by the LiDAR solution of Horizon Journey 5 (128TOPS computing power chip); Perception hardware, composed of 10 cameras + 12 ultrasonic radars + 1 millimeter-wave radar.
In this way, the optimization direction is relatively simple, because the number of hardware involved is less, three cameras + two radars. Combined with the above functions and optimization strategies, it is to improve the accuracy and decision-making weight of "long-distance perception". Among this set of perception equipment, millimeter-wave radar has the farthest sensing distance, followed by cameras and ultrasonic radar; The sensing distance of the former is about 160 meters, while the sensing distance of the camera will be slightly shorter than that of millimeter-wave radar. These two sensing hardware make up the vehicle's recognition and perception of forward-facing objects.
In the case of only 128TOPS computing power, there are not many complex instructions that can be done, so in order to reduce the game time of forward perception data brought by millimeter-wave radar and high-definition camera, the strategy should be to adopt a combination of two data, the object picture of the camera + the object position of the millimeter-wave radar to make the final decision, and the priority of long-distance objects (there should be distance limit trigger conditions) will give more decision-making weight to millimeter-wave radar.
Under the premise that the hardware is not adjusted, the optimization of the algorithm strategy requires a relatively large adjustment. In particular, ideal intelligent driving (NOA working conditions) has encountered the following concentrated situations: recently, the high-speed misidentification of the vehicle on the billboard caused by false braking; In rainy conditions, the perception camera mistakenly identifies that someone is chasing after the vehicle.
All of the above are caused by bugs in the graphics algorithm, so after this OTA, there is a reason to continue to optimize the graphics algorithm significantly, and then test this set of NOA functions. For the time being, the 1320km takeover is likely to be achieved under the premise of sunny daytime.
After catching up with Huawei, can it surpass?
After looking at this OTA first, the ideal user experience needs to be changed, including drawing dragons on relatively wide lanes (testing back and forth at the edge of the lane line), and occasionally getting too close to the left guardrail; In the case of narrowing lanes, dual lanes and single lanes, etc., the coping ability is not enough, and there will be takeovers. Then, when there is a vehicle in front of the vehicle that intrudes into the lane, the braking sensation is slightly heavier.
These are all high-speed NOA usage in version 5.1, but the road test distance is not long, about 100km.
In addition, through the AD Pro 3.0 (version 5.2) on the inside, it can be seen that some of the above problems have been modified, such as the drawing of the dragon in the lane marking, and the heavy feeling of braking in the lane invasion. However, there are still some problems, for example, it is more dependent on the objective condition of speed limit, and it is not possible to intelligently judge how fast the vehicle needs to travel more appropriately. In addition, in the case of lane change, a certain amount of deceleration is required before the lane change can be performed, and the efficiency of lane change (continuous lane change) is relatively low; Similarly, after a failed lane change, it becomes indecisive to change lanes again.
So, can it surpass Huawei?
If we only consider the takeover rate from one dimension, the takeover rate of the ideal and the question is about the same, both of which are at the level of about 1300km and 1 takeover. However, there will still be a slight gap in the use experience. The sense of experience is that ADS 2.0 is more stable under high-speed NOA conditions.
For example, on the previous Wenjie M7, whether it is a straight line, lane change, cornering, or even a sharp curve under high-speed NOA conditions, the performance of the steering wheel is relatively stable, and there are few small movements (or there are, but it is more difficult to detect). For this kind of processing in details, it is more of a combination of perception + vehicle control, and the processing of details is better. After that, under the condition that the forward lane is suddenly encroached upon, there is no obvious pause in deceleration, and it is not abrupt, which is also a better point, at least in terms of driving experience.
However, it should be noted that this set of perception hardware is matched with lidar. Therefore, it is better to grasp the details, and it is also necessary to admit that the optimization of algorithms and strategies is good, and the products with lidar can use high-speed NOA very stably, and there are relatively few of them.
As for the case of false braking, the ADS 2.0 system is the same as that of AD Pro. Some appear in the car-free lane, and some will appear after the corner, there is no law at all, which may be caused by errors in the perception data, which needs to be optimized later.
The current answer is that it has not achieved better handling of details, such as the stability of driving, the degree of handling to deal with lane intrusions, and the strategy of changing lanes. However, without the addition of LiDAR hardware, relying on the combination of camera + radar, it is also worth affirming that about 90% of the experience of ADS 2.0 can be achieved.
In the end, the purpose of car companies in rolling high-speed NOA is to reduce the fatigue of the driver on the road, so that the car can run on its own in the high-speed section better and smoother, and try to take over as little as possible. Only after this premise is done, there is a reason to optimize the detailed experience, and now it is ideal to strive to do a good job in the ability of high-speed NOA "Shun"; Huawei's ADS 2.0 is heading for the stage of "easy to use".