The ultimate goal of most car companies is to achieve automatic driving, although some cars have been equipped with auxiliary driving functions, but still can not leave the driver to operate, that is to say, the current car technology is not developed enough, still can not achieve the real automatic driving function, but has achieved the basic automatic driving function, such as automatic parking function.
And now new energy vehicles than the traditional fuel vehicle configuration is higher, especially the technology configuration, according to most of the car companies' research shows that in order to achieve the car automatic driving function, the need for LiDAR (light detection and ranging) sensor, lidar working principle is to use the laser to 3D analysis of the surrounding environment, by measuring the time required to emit laser pulses and return from the target to calculate the distance and three-dimensional effect, so as to achieve the function of automatic driving.
That is to say, the speed of the sensor returning from the target is fast enough to allow the computer in the car to react quickly, and this reaction speed must reach the human brain reaction speed, or even exceed the human brain reaction speed, and must be accurate and error-free technology, many car companies believe that this is the key to the automatic driving function.
In the early days, Mercedes-Benz cooperated with lidar companies and confirmed the safety and accuracy of lidar technology, and the model equipped with automatic driving technology is S-class, configured as L3 automatic driving system, in addition to Mercedes-Benz, most new energy vehicles are also equipped with lidar automatic driving technology, and recognize the safety of lidar technology to achieve automotive automatic driving function.
As the leader of new energy vehicles, Tesla can not be uninterested in the automatic driving function, and Tesla's autopilot technology is different from other car companies, Tesla did not use lidar technology, Tesla feels that lidar technology is not the ultimate choice for automatic driving technology, but in the radar premise to add camera technology, that is, equipped with cameras, and is the whole car 360 degrees without dead angles to shoot the surrounding environment.
This technology works by accurately obtaining information about the surrounding environment to determine the direction and speed of travel, so Tesla insists on using cameras and radar to achieve automatic driving technology, in addition, Tesla also abolished the previous lidar technology and announced the use of the "Tesla Vision" system that relies on cameras, but Tesla's unique choice has attracted attention from all walks of life.
At present, if you want to achieve full commercialization of automatic driving, the premise is that you must ensure safety most reliably, so many people are worried and questioned about Tesla's "visual only rely on camera" method, because under normal circumstances, the camera is more susceptible to the weather environment than the lidar pulse, that is, the camera is not as practical as lidar, so why tesla will still adhere to the camera autopilot function?
Most car companies use lidar, why does Tesla insist on using cameras?
Lidar vs visual perception performance competition
Lidar perception technology is led by lidar, supplemented by millimeter-wave radar, ultrasonic sensors, and cameras. The working principle of lidar sensing the environment is to emit a laser beam through lidar, measure the time difference and phase difference of the laser in the process of emission and recovery, determine the relative distance between the car and the object, and realize the real-time perception of the environment and obstacle avoidance.
Lidar has a long detection distance and high accuracy, strong anti-jamming ability, can actively detect the surrounding multi-object environment, obtain the surrounding environment point cloud to build a 3D environment model. Even if the light is not good at night, it will not affect the detection effect. Although lidar is not afraid of dark light, it is sensitive to weather, rain and snow, sand and dust, fog weather and so on affect the lidar recognition effect. The lidar fusion high-precision map scheme can effectively make up for the defects of high dependence on the visual solution environment and large demand for computing power, and its performance advantages make most automakers list lidar as an indispensable sensing device for L3 level and above automatic driving.
Visual perception is a camera-led solution, and the cost of the camera is extremely advantageous over lidar. The price of the camera is around tens of dollars, while the lidar is in the hundreds of dollars, which is several times the price. In addition, camera technology is gradually mature, high-resolution, high-frame rate imaging technology makes the perceived environmental information more abundant, but the camera is limited in the dark environment, and the accuracy and safety have declined.
For example, Tesla's most criticized ghost brake failure is that in the shadows of some tunnels and bridges, because of the structure of the camera, the algorithm will suddenly appear in the shadow as an obstacle, causing the vehicle to suddenly automatically slow down, causing safety hazards. In the visual solution, the hardware performance is compared, and the camera function is slagged by the second. Thanks to the addition of software algorithms, visual solutions can rely on powerful algorithms to ensure the normal operation of image processing and decision execution.
Compared with lidar, the weakness of visual perception is more obvious: the camera relies on light conditions, the accuracy of the perception method is low, the dependence and requirements on the algorithm and computing power are extremely high, and the barriers to data acquisition and algorithm iteration are high. In terms of performance, lidar obviously wins, Tesla spends a huge cost on computing power and algorithms, the investment is not small, the head iron has been standing up to adhere to the visual perception route, in the end what are the angles of consideration?
Tesla focuses on pure visual route logic
In Musk's view, "pure visual perception is the road to real-world AI", and this is also the underlying idea he pursues to solve the problem - the first principle, that is, to return to the most basic conditions of things, split them into elements for structural analysis, so as to find the best path to achieve the goal.
In the process of driving a vehicle, we collect road condition information through the eyes and supplement it with brain processing, so that automatic driving can theoretically also be safely driven through visual perception supplemented by algorithmic processing.
What Tesla wants to do is mimic the ability of human vision to obtain information to achieve autonomous driving. Since the perception method of the visual camera is less accurate, it relies on Tesla's unique data advantages and the ability to build computing power and algorithms to smooth out this defect.
In terms of data, while other self-driving manufacturers are still collecting data in the road testing stage, Tesla has accumulated a huge amount of real road condition data thanks to the millions of cars with cameras sold around the world. The data used for deep learning model training has long established barriers for Tesla's algorithm, and the accumulation speed of these data samples and the efficiency of the algorithm cannot be copied by other manufacturers, and can only be blinded and anxious.
In terms of computing power, Tesla's newly established supercomputer Dojo has a powerful computing power, which is set up for Tesla's autopilot system to concentrate on training the entire autopilot system.
At the technical level of the camera, Tesla has also carried out technological innovation, using "pseudo-lidar" technology instead, the depth estimation of pixels in the camera, similar to the point cloud function of lidar generally forms 3D target detection, improving the accuracy of depth estimation, and the gap between lidar and camera begins to narrow.
People rely on vision when driving vehicles, and our neural network can process signals such as distance and speed in visual information, and Tesla's neural network seems to be able to do it gradually. Tesla's visual perception route is gradually narrowing the gap with the lidar solution, but the price behind it makes it impossible for latecomers to follow and copy, which also establishes a strong barrier for Tesla. The pure visual scheme is supported by massive sample data training learning and advanced image processing algorithm computing power, which is destined to be a difficult route chosen by a small number of climbers.
epilogue
Tesla has embarked on the road of pure visual FSD, although there are some accidents, but the current autopilot experience is generally more advanced, there is no doubt that Tesla will continue on this pure visual FSD road.
Of course, the automatic driving route using lidar sensors has more manufacturers and more customers and consumer groups besides Tesla, and it is impossible to prove who can be better by using pure visual FSD or using a variety of sensor fusion.
Perhaps, this is the same as the Apple iOS and Android camps in the smartphone market, and eventually both have their own markets.
And in the market, there is no second Apple, but there are countless Android manufacturers.
Source: Neman Motors
E N D