Text/Qin Ming
At present, there are two large technical differences in the field of sensing technology in the field of automatic driving, one is the lidar scheme, and the other is the camera, whether it is a traditional car company or a new car-making force, it basically uses lidar perception, only Tesla is still adhering to the method of pure visual sensing of the camera.
Camera sensor
Tesla's philosophy is that since people can observe the surrounding environment by looking at the surrounding environment by both eyes, the autopilot system can also use the same logic to rely on the camera, so Tesla's camera is imitating the human eye.
The camera can pass the most realistic road surface image to the processor, and then convert these images into data, the most difficult of which is not how to restore the real image, but how to convert the image into data, which has a very high demand on the processing and storage capacity of the data.
The visual solution has a relatively high dependence on sample data, and it needs to be trained with a large amount of data, continuous optimization, and continuous learning, so Tesla has long built a closed-loop framework for data collection and learning, collecting various driving scene data through the camera, and uploading it to the Tesla cloud, forming a real and huge data pool.
This cloud-based database can provide learning templates for Tesla's intelligent algorithms, giving the algorithm the ability to continuously upgrade. Based on the collected driving data, Tesla will continuously optimize the algorithm through "shadow mode", and finally upgrade the OTA of each Tesla FSD autonomous driving.
Lidar sensors
Lidar detects and measures distance through a laser beam and consists of a laser transmitter, an optical receiver, and an information processing system.
Lidar can quickly reconstruct the three-dimensional model of the target and various pieces of data by emitting and receiving the laser beam, analyzing the turnaround time after the laser encounters the target object, calculating the relative distance between the target object and the car, and using the three-dimensional coordinates and reflectivity of a large number of dense points on the surface of the target object to quickly reconstruct the three-dimensional model of the target and various piece data, establish a three-dimensional point cloud map, and draw an environmental map to achieve the purpose of environmental perception.
Since the ranging formulas are fixed, these algorithms are naturally the same, so the improvement of lidar lies in the accuracy of ranging, which is completely dependent on hardware performance and cannot be improved through self-learning.
The differences and characteristics of the two
Lidar is an active sensor, through the emission of pulsed laser and detection of the target's scattered light characteristics to obtain the depth information of the target, with high accuracy, large range, strong anti-jamming ability characteristics, can be in some extreme weather and night to achieve higher accuracy than the camera, to prevent the vehicle's misjudgment, improve the redundancy of safety.
However, the data obtained by lidar is sparse and disorderly, difficult to use directly, and the characteristics of laser monochrome make it impossible to obtain color and texture information, although the ability to map the distance of the surrounding environment is outstanding, in some roads without features, such as road sections without centerline signs, lidar will be difficult to distinguish or even miss the problem, so it must be used with other sensors complementarily.
In contrast, although the data obtained by the camera is more similar to the real world perceived by the human eye, there are light hardware and heavy software characteristics, the field of view has limitations, when driving to see things can only see the front and side, the judgment of the situation is not completely correct, and the field of view is affected by the environment is very large, and the difficulty of completing image acquisition in a harsh environment will be greatly improved.
In addition, because the vision scheme in the context of reduced hardware requirements, its requirements for software are significantly improved, that is, it needs to rely on powerful algorithms to ensure the efficiency of image processing, command issuance and processing.
Why does Tesla insist on using cameras as sensors?
With so many limitations on cameras, why has Tesla been embracing visual sensing? Many people can think of a reason, that is, to reduce costs.
Starting with Autopilot 2.0 and now FSD 10.6 Beta, Tesla models have stuck with 8 cameras. At present, the price of a camera is expected to be around $10. The cost of lidar has also been reduced to hundreds of dollars, and the current lidar model vehicles are generally equipped with a long-range lidar on the roof, and a short-range lidar is equipped on each side, and the price is about 600-800 US dollars.
Therefore, Tesla can harvest about 5000 cost price, which is indeed a factor, but personally, the main reason is that the ultimate purpose is different, Tesla uses cameras to measure, can detect the surrounding environment, when Tesla's car can be driven to every corner and angle, then Tesla has a very terrible database, in the current information age, these data is the most valuable wealth.
Tesla wants to achieve the real unmanned driving function, so the pursuit of the same evolutionary ability as humans, unlike lidar limited by the algorithm fixed, the camera can determine whether the obstacle is fixed or active, just like an animal on the road, the camera plus cloud data can clearly identify the match, and deduce that the animal may move to decelerate and change direction in advance.
Write at the end:
In general, lidar is just a ranging device, the camera is more like our eyes, and the content offered is richer, but sometimes the eyes also deceive us, both have their own advantages, but also have their own obstacles to development, in any case, both technologies are expected to achieve true autonomous driving.