Recently, Xpeng Motors announced that it will be in? 202 1? The model produced in 2000 was equipped with lidar, which caused the dissatisfaction of Musk, a senior "black powder" of lidar: "They have the old version of Tesla-related software, but they don't have our latest neural network computer." Musk also stressed that this view is only for Xpeng Motors.
Obviously, Musk Tucki is not only the technical route dispute of lidar, but also Tucki's personal emotion of "copying" Tesla's autopilot code. However, whether Tucki stole the Tesla code still needs to be decided by the court.
Under the deadlock that both parties refuse to publish their own codes for comparison, the "patent dispute" of Tucki Tesla will continue for a long time, and it is very likely that it will come to nothing in the end.
Let's put aside old grievances for the time being and talk about the immediate things first.
After Musk tweeted about Tucki, He Xiaopeng, the founder of Xpeng Motors, was not idle on social media. He Xiaopeng fired directly: "Yesterday, it seems that Tucki's next-generation autonomous driving architecture, including lidar, was released, which made some people in the West very unhappy and used pigu to make continuous voices. What I want to say is that rumors have long proved that you can't beat any competitors. From next year, you should be prepared to be defeated by us in China. As for the international community, we will meet. "
There are two key points in this article. One is to reiterate that "stealing code" is a rumor, and the other is to threaten that your own autonomous driving will defeat Tesla? Fire Services Department.
Where did he get Xiao Peng's confidence? First, Tucki's latest navigation system? NGP？ Excellent performance in the small-scale tests of media and users, and second, the application of lidar next year will be a "tiger's wing", which gives him considerable confidence.
The dispute over the route of autonomous driving sensors is a cliche in the industry. Tesla is a loyal visual party. Musk once said that "everyone who uses lidar is a fool", but the voice in the industry is not as extreme as Musk's. More voices think that "the scheme of combining lidar with camera is the safest". After all, lidar is dealing with bad weather and? 3D？ The precise positioning of objects in space is much better than that of cameras.
Although everyone thinks of lidar, but in addition to some test cars that study autonomous driving startups, is there an Audi that "only hears its voice but doesn't see its car"? A8？ Moreover, lidar has basically not been put into mass production. In the final analysis, the reason is still too expensive.
However, Xin Huang, Director of Autopilot Products of Xpeng Motors, said: "At present, the requirements of vehicle regulations have been solved, and the cost is good. It is not as outrageous as expected, and it is still within the acceptable range." As to whose lidar to use, Tucki has not reached a conclusion, only knowing that several suppliers are competing for products. Considering the cost, these companies must focus on domestic brands.
The effect of lidar on trains will not be tested until next year, but? NGP？ We have already experienced it. Did he brag in Xiao Peng? Combining my previous test drive? Models? 3? Experience, I made a subjective comparison between the two pilot systems.
Look at the sensor configuration first, Tucki? P7？ what's up 14 sensing camera, 5 millimeter-wave radars, 12? Especially ultrasonic radar.
1 front three-eye camera
1 Front monocular camera?
5 enhanced sensing cameras
4 panoramic cameras
1 Driver Monitoring Camera
Fifth Bosch Fifth Generation Millimeter Wave Radar
12 ultrasonic sensor
Tesla? Models? 3? It is:
1 front three-eye camera
5 panoramic cameras
1 front-end millimeter wave radar
12 ultrasonic sensor
1 car camera (not enabled)
Judging from the data recorded by the sensor, Tucki? P7？ Is it significantly better than Tesla? Models? 3? Yes But what about Tucki? P7？ Carrying NVIDIA? Xavier? Is the chip computing power slightly inferior to Tesla? Models? 3? what's up FSD？ Chip. There is another essential difference, and that is Tucki? NGP？ With the high-precision map support of Gaode, what about Tesla? NOA？ It depends entirely on bicycle intelligence. The application scenarios of the other two systems are also different, Tesla? Noah? Only highways are supported; Tucki? NGP？ Besides expressways, it also supports some urban roads.
In terms of configuration, Tesla is worse than Tucki, especially the richness of millimeter-wave radar and the lack of high-precision maps. Looking at the configuration alone is definitely one-sided, and we speak with the experience of actual road running.
Tucki? NGP？ The test environment is chosen along the Nansha-Humen Expressway in Guangzhou, where the traffic volume is very large, especially the large trucks are in droves, which poses a great challenge to the driver assistance system.
The pilot system tests nothing more than three functions. The first is to automatically change lanes, automatically enter the ramp, and identify ice cream cones.
When the system detects that it is suitable for opening? NGP？ Time, dialed twice in a row, NGP? Officially opened. This has nothing to do with Tesla? Noah? The logic is similar, but the opening moment, NGP? The experience is not good enough, and the system intervention is blunt. Even if the line is not pressed, the steering wheel will be pulled very hard to completely center the vehicle. Although this action is not dangerous, it will make the driver's heart tremble for the first time.
At this point, NoA? Deal with it better.
Activate? NGP？ After that, the vehicle can be stably kept in the lane. On the regular straight road section, NGP? And then what? Noah? The performance of the two is equivalent. When encountering ups and downs, is there a high-precision map blessing? NGP？ Will behave more stably.
Automatic lane change, NGP? Can independently complete a series of operations such as automatic lighting and automatic lane change. The logic of active lane change is this: within the speed limit range, when the current vehicle has a certain impact on the cruising speed of Tucki or needs to get off the ramp, the vehicle will consider lane change. The whole body sensor of the vehicle? 360 perception will detect whether there is a threat in the adjacent lanes. Under the condition of ensuring safety, the system lights up and is ready to change lanes.
Drivers with different habits will have different feelings about the waiting time from lighting to changing lanes. Some colleagues think that this time is too long and the system is not decisive enough. I think this setting of the system is understandable. Traffic regulations require that the lights should be turned on for more than three seconds before changing lanes, but many of us can't do it in driving habits, so this forms a contrast.
Different from human drivers, there are on-ramp and off-ramp Does China Expressway ramp often speed limit? 60km/h？ What's the minimum? 30km/h, 40km/h, but humans generally decide their own speed according to road conditions, but machines don't. When the manufacturer designs the program, everything is still based on laws and regulations, which leads to the disconnection between the machine operation and the actual driving habits of human beings.
Although this separation will affect the driving experience of many "old drivers" to a certain extent, it is always right to drive according to the law. If traffic accidents are caused by the "radical" system, the final division of responsibilities will be more "chaotic". However, this conservatism has also brought some disadvantages. Because large trucks often occupy the right lane densely, this terrible moving roadblock often makes many cars miss the opportunity to get off the ramp, and human drivers still need to "intrigue". If the machine continues to hesitate, it will be easier to miss the opportunity. Of course, it's all a question of probability. Facing the heavy traffic, Tucki? P7？ Go down the ramp smoothly every time.
Although the overall lane change stability is not bad, what about Tucki? NGP？ It's not without slots. For example, when changing lanes, vehicles in the third lane want to merge with you in the same lane. At this time, it is difficult for Tucki to make accurate perception, which is also the difficulty of many driver assistance systems, and it is not an easy task for human drivers.
Through practical experience, I come to the conclusion that Tucki? NGP？ Autonomous lane change is generally better than Tesla? Noah? Even better, on the one hand, more abundant sensors make Tucki sharper and more confident in making decisions. On the other hand, what about Tucki? NGP？ The algorithmic ability of behavior prediction needs to know more about the driving habits of drivers in China, so as to make more accurate behavior planning when encountering obstacles in front.
With the help of high-precision maps, Tucki? NGP？ Being able to calmly deal with curves with various curvatures when entering and leaving the ramp is indeed much more stable than Tesla.
However, this function is not perfect enough. We had two "panic moments" on the same ramp at different times. In China, many turnouts merging into expressways are funnel-shaped, that is, multiple lanes are merged into one or two. It was this slope that tripped Tucki. In the test drive. The Tucki we drove? P7？ When it is about to merge into the main road, the traffic on the right rear side suddenly tends to be congested. P7？ Quickly made a braking action, and the speed dropped to? 20km/h？ Even lower avoidance almost led to rear-end collision.
The two drivers in our car, including me, agreed that there would be no such emergency when driving. When there are no obstacles ahead and the traffic jam has not come to occupy the driveway, we just need to speed up and pass quickly to avoid embarrassment.
I consulted Xin Huang, Director of Autopilot Products of Xpeng Motors, and Xin Huang said, NGP? When there is traffic jam on the right rear, the strategy is to slow down and avoid. Engineers believe that in traffic accidents, the harm caused by low speed is always less than that caused by high speed.
In addition, I personally feel that if we deal with similar scenes in the way of human drivers speeding through, we need the system to make more predictions about threatening vehicles. What is the opponent's speed? How far is the other side? What speed should we accelerate to? After all, this is a sudden rear car, which is difficult to perceive. It is not easy to make an accurate prediction. The safest way is to slow down and make way. This is the vast majority at this stage? Adas? The usual strategy of system and autopilot system: don't provoke without provocation, just make it safe.
Humen Bridge, with huge traffic, has always been a "construction major", and there are always countless ice cream cones here. Compared with general construction roads, Humen Bridge is simply an "additional problem". Tucki has no problem in identifying ice cream cones, but sometimes the unreasonable placement of ice cream cones really makes Tucki? NGP？ A bag, when the ice cream cone tilts slightly in this alley, NGP? Will you show obvious self-confidence, especially when there is heavy traffic in front, or? NGP？ Exit, otherwise the system will automatically slow down to an unreasonable speed. In short, the experience needs to be strengthened.
You have to go through it, Tucki? NGP？ Really better than Tesla? NoA, this is also the unanimous answer of my colleagues in the media. For the sake of understanding, let me briefly summarize Tucki? NGP？ Better than Tesla? Noah? Excellent place, which is largely brought by the advantages of sensors and high-precision maps.
1、NGP？ Behavior prediction and behavior planning are more reasonable, and there are few unnecessary lane changes.
2. Facing the ramp with large curvature, NGP? More confident
3. There is a reasonable plan for the time to get down the ramp.
Tucki? NGP？ Insufficient:
1, facing the lack of perception of traffic jam on the right side, blindly letting the speed be easy to rear-end.
2. The high-precision map data is temporarily unstable, and some entity roads are not accurately identified.
3. Dealing with the complicated ice cream cone scene is still unstable.
Although Tucki as a whole? NGP？ BitTesla? Noah? Better, but Xiao Peng said that Tucki's autonomous driving will beat Tesla, which is not easy to achieve. Tucki's advantages are the existing sensors, car-following lidar, high-precision maps and autonomous driving algorithms, all of which are more familiar with the habits of China consumers. What are Tesla's core strengths? It is a shadow mode, a neural network computer in Musk's mouth. Tesla's existing market scale is much larger than that of Tucki, and the data collected by these Tesla people who run on the road every day will be continuously fed back to Tesla's autonomous driving. The great power of deep learning will make up for the shortcomings of many algorithms and hardware, and Tesla's visual technology and algorithms will become more and more sophisticated. Although lidar has inherent advantages, this advantage will be shortened.
As the first company to mass-produce the pilot system, Tesla's technical reserves should not be underestimated. Although Tucki has made temporary achievements, it is not easy to leave the old-timers behind at once. He Xiaopeng said that it is a "angry talk" with certain emotions to hit Tesla to the East, and he does not rule out the suspicion of "self-hype". Judging from the current technical performance of Tucki, this system has great potential, and Tucki is qualified and capable to compete with Tesla in a higher level of autonomous driving competition.
This article comes from car home, the author of the car manufacturer, and does not represent car home's position.