Lidar (LiDAR), which uses visible light pulse reflection to accurately measure object distance, combined with radar and camera is considered the best development path to achieve full self-driving. Tesla, who disagreed, argued that as long as computer vision technology is good enough, lidar is actually a redundant device. After all, humans can drive by eye.
The corresponding solutions are naturally different. Tesla, which does not use the Lidar plan in the future, mainly uses 8 photographic lenses (3 front, 2 side and 1 rear), and front-end radar and some ultrasonic sensors for close-range detection. Other vendors tend to add one or more lidars, even with night vision lenses to improve safety.
Tesla's method costs less, but requires more data processing power. For manufacturers who want to introduce self-driving as soon as possible, the safety considerations are higher than the cost, so choose the option of using laser radar plus camera.
Reliability is the key. The camera can only guess the distance by sensing the relative size of the object, and must adapt to the changing light. Unlike Lidar, the distance can be measured accurately day and night.
Although computer vision technology has made major breakthroughs in the past few years, it is still not perfect. Tesla believes that humans can drive with their eyes alone, so it is believed that computer vision can definitely achieve this capability. Its CEO Elon Musk believes that once computer vision problems are solved, there is no value in laser radar. However, the question is when will this day come? Other manufacturers choose lidar because the speed of laser radar price reduction is faster than computer vision technology breakthrough.
A recent paper published by Cornell University proposed a method to increase the accuracy of target detection within 30 meters to 70%, showing that computer vision systems are challenging the core values of laser radar reliability and correct distance measurement. However, the depth accuracy of the lidar in the range of 100 to 300 meters is close to 100%, which is very important for highway driving, and 30 meters is not enough for urban driving.
In summary, Musk believes that super-strong and near-perfect computer vision can achieve the goal of safe driving, making Lidar useless. Others believe that driving reliability can be achieved by combining the reliability of laser radar with good enough computer vision technology.
Even if Tesla can't solve the problem of full self-driving with hardware, at least it can maintain its electric vehicle business with its own advantages in electronic power transmission and high-tech car design. Electric vehicles that are not equipped with lidar can still automatically park in the parking lot, cope with traffic congestion, and reduce the possibility of collision through a better advanced driver assistance system.
Lisa supporters also use neural networks to process lidar data like parsing camera and radar data, and because of the lower resolution of lidars, the neural network architecture is simpler and faster.
The ultrasonic sensor can reach a distance of 5 meters. Under ideal conditions, the distance measurement can reach 20 meters, and it has high measurement accuracy, which will greatly improve the safety of driving, and the low price is more beneficial to the amount of traffic. Production. Compared with the current two-meter ranging, the 20-meter range is already 10 times larger, and most importantly, it has high accuracy in such a large measurement range. The material used in the ultrasonic sensor is a new type of piezoelectric ceramic material, and the signal processing technology adopts multi-scene signal processing and analysis technology under complex medium environment conditions, and the precision can reach centimeter level.