[Aika Auto Technology Anecdote Original]
In 2021, the hot words of the car circle are bound to be indispensable to lidar, and some people call this year the "first year of lidar". The reason for this is that this year, not only a number of new cars on the market are equipped with lidar, but the price of models equipped with lidar has also dropped to the price range of 200,000 yuan. In short, at your fingertips.
As the saying goes, people are not red, and lidar is also controversial in the circle. More famously, Tesla founder Musk has publicly attacked lidar many times, including but not limited to doomed failure, stupid things...
It can be seen that although lidar has fans, there are also industry giants that are strongly opposed. It seems that today's lidar is like an automatic transmission with a turbocharged engine yesterday, some people are optimistic and some people are looking down.
The essence is the clash of perception strategies
Autonomous driving requires the vehicle to have a clear understanding of the external environment, and then perform corresponding response actions, such as obstacles in front of it, and emergency braking after the vehicle perceives. At the level of response strategy, everyone is similar, and the key point is how to perceive the environment clearly, in a timely and complete manner.
Around lidar, we can divide today's autonomous driving technology route into two factions, namely the "pure vision system" that uses cameras, and the "composite perception system" of millimeter-wave radar, lidar, cameras, and high-precision map combination applications.
The pure vision system is represented by Tesla. Musk believes that since people rely on their eyes to complete driving, why can't cars work? This statement is really quite reasonable, according to Musk's logic, equipping the car with enough visual perception ability is theoretically no problem.
However, under this technical route, Tesla must have a powerful enough algorithm to interpret the information in the picture. After all, the camera collects a 2D picture, which is difficult to restore the distance in the real scene, so there will be multiple cameras to collect the picture from different angles and build a 3D scene. In addition, the camera will also be affected by the lighting environment, for details, please refer to tesla and white truck accidents.
It is also very important that the pure visual scheme will also throw a difficult problem to the algorithm, give the system a map, and leave it to the system to judge. Therefore, algorithms that work with pure vision schemes require a process of learning to recognize objects.
Pure vision schemes also have advantages, and are obvious. First, cameras can identify traffic lights, road signs, lane lines, signs, etc. Secondly, it is also very important that the cost of the camera is relatively more controllable.
In the department of compound perception, lidar is mainly talked about. Lidar is a lidar, through the laser to reach the reflective surface after returning to the light source to complete the measurement of distance, in addition to the basic function of ranging, lidar can also be detected by detecting the contour, position, volume and other information to draw a 3D model of the surrounding environment. As a lidar, bad weather such as smog and sandstorms are the natural enemies of lidar.
Therefore, lidar is mostly used in combination with millimeter-wave radar, and millimeter-wave radar has a strong ability to penetrate smoke and dust, and can be competent for night work.
Stage achievements of the two major factions
It is obviously untenable to talk about theory without talking about the results.
In March, Honda received approval from Japan's Ministry of Transport, and next March the autonomous driving system will appear on honda's Legend model, the world's first L3 production car to be officially approved. The car is equipped with a binocular camera, 5 millimeter wave radars, and 5 lidars.
Not surprisingly, the price of this car is very expensive, 7.249 million yen, about 433,000 yuan. Some insiders have disclosed that the cost of this car in terms of sensors is as high as millions of yen. Don't forget, this does not include the software and hardware costs related to the hardware of the perception system, such as algorithms and chips.
Seeing this, you may understand why Musk would think lidar is a stupid solution for autonomous driving. Because the cost is too high, it is really difficult to promote on a large scale.
So where has Tesla, as the head of the pure vision department, progressed? Unfortunately, Tesla did not become the first manufacturer to have an officially certified L3, nor did it become the second. The only achievement worthy of recognition should be affirmed by the American Highway Safety Insurance Association (IIHS) in the safety test. IIHS told Tesla Model Y that you did a good job of pedestrian collision warning. Tesla still can't get to L3, and IIHS doesn't care about it.
On the one hand, there is some substantial progress, but there is still a long way to go before the ultimate goal, and the other side has not yet made any breakthrough.
Who is more reliable
Honestly, this reminds me of the "Cold War." The two sides have different ideas and ideas, but both believe that they are the leaders of the future of human society. Most people could not have imagined that the devastated giants after the war could go low and high, and most people could not imagine the Christmas gift after the Soviet Union attacked and defended the United States.
The same is true of the route of autonomous driving technology, and the result is not known until the end, and perhaps the final result is that all roads lead to Rome, and everyone can not say that they have the same destination. But rational analysis of the process and original intention, we can still get some answers.
What was the original intent of autonomous driving? safe. What does security require? redundancy. The camera can indeed do a lot of things, and the complete body of the algorithm can also bring unexpected surprises, but if the existence of millimeter-wave radar and lidar is not dragged down from the cost, is it more reassuring? For example, the door of the home has an anti-theft lock, we will be at ease, if the door is connected to the property security system, there is a strong demolition behavior will immediately call the police, will we be more at ease?
In the past few years, with the automatic driving on the wind, the cost of lidar has dropped steadily. Previously, the cost of lidar used in Google's self-driving car was about $75,000, and then Waymo reduced the cost to $7,500, and this year, a domestic self-driving solution provider announced that it would reduce the cost of lidar to $200. Regardless of whether it can be achieved or not, at least reducing costs is the general trend of lidar.
The Great Wall Mech Dragon equipped with four lidars is priced at 488,000 yuan
Now, we can see lidar from a small number of cars. Although its arrival has not helped the vehicle to leapfrog from L2 to L3, from another point of view, the application of mechanism radar helps to improve the perception level of the vehicle, which is a topic that cannot be bypassed to achieve higher-order automatic driving. Not only limited to lidar, compound perception is obviously more in line with redundancy in terms of safety, and more directly hit the ideal of automatic driving, but there are still some problems that need to be overcome by multiple industries together.
Who is more reliable---- 1