"Because we need to think about how to sell things. So first of all, we will consider the relatively low price of laser pointer." This is the idea of Hou Xiaodi, CTO of Tucson. In his view, lidar has not yet been mass-produced, and the price is expensive. At present, it cannot generate enough value, but Tucson does not rule out choosing to use lidar after the price reduction.
However, as academician Gao Wen said, it is so difficult to reduce the cost of lidar. When the price can be reduced to meet the mass production requirements and accepted by most companies, no one knows. On the other hand, as companies in the field of autonomous driving compete to advance the process of mass production of their products. There is not much time left for these companies. Rather than all of them are "killing" in Lidar technology. It is better to look for other technologies that are more feasible and likely to land in the short term.
Visual perception is a technical route that can solve problems with low cost and big data. Although compared to green laser pointer radar. There are many advantages that make it easier to promote the commercialization of autonomous vehicles. But this route itself is currently facing some technical difficulties. "If you want to use visual perception to handle navigation problems well, the algorithm of AI decision is very important." Academician Gao Wen said. "Taking visual distance measurement as an example, after the vision camera inputs some environmental images, through related calculations, we can reverse the distance between the car and the house, pedestrians, and signal lights. So the algorithm for distance measurement must be good enough here. To be accurate and accurate. "
However, in actual working conditions, the phenomenon of position drift often occurs when using visual ranging. Because visual ranging is based on the camera to obtain different environmental images, and then compare with the original map. And through related algorithms, calculate the distance. during this process. If the data processing speed cannot keep up, or the algorithm is not good enough, drift will occur. There is a certain gap with the actual results.